LISTEN TO THIS ARTICLE

For the past 2 years, the fate of a major plant collection has worried scientists. Duke University shocked botanists when it announced in February 2024 that it would close its 100-year-old herbarium, citing the cost of upgrading crowded, aging infrastructure. Researchers rely on herbarium collections to study plant evolution and biodiversity, as well as the effects of climate change.

This week, the University of North Carolina (UNC) at Chapel Hill—which is located about 16 kilometers from Duke—said it will incorporate Duke’s 850,000 plant specimens into its own herbarium . “This combined collection will be a powerful resource for studying and conserving plant diversity in the Southeast,” says Pamela Soltis, a plant evolutionary biologist at the Florida Museum of Natural History.

Together, the 1.7 million specimens will create a collection only exceeded in size in the United States by those at Harvard University and the University of California, Berkeley. Next year, the specimens will be moved into two empty floors of a UNC building. The long-term plan is for both collections to be housed in a new plant biodiversity research building at the North Carolina Botanical Garden, located nearby.

UNC has been thinking about constructing the new facility since at least 2016, when it spent $250,000 on design work . Whether it is built, however, will depend on fundraising and budget allocations from the state.

“I’m pleased the Duke collection will remain intact and in North Carolina,” says Kathleen Pryer, former director of Duke’s herbarium. “UNC has saved the collection. Now it needs to finish the job: Build the scientific program that makes this the third-largest university research herbarium in the country, not the third-largest pile of specimens.”

LISTEN TO THIS ARTICLE

Senator Bernie Sanders (I–VT) yesterday announced he plans to introduce legislation that would ban artificial intelligence (AI) systems that achieve “superintelligence.” But the move has revived an awkward question that has long divided AI researchers: What counts as superintelligence?

“I certainly take issue with any legislation that centers around the concept of ‘superintelligence,’” says Heidy Khlaaf, chief AI scientist at the AI Now Institute. “Not only does it not have a consistent definition, but it’s often hypothetical and unfalsifiable.”

Sanders said he planned to join with Representative Greg Casar (D–TX) to introduce the Ban Artificial Superintelligence Act, which would permanently prohibit the development of AI systems that exceed human cognitive abilities. “Despite its potential deadly consequences, cutting-edge AI technology is less regulated than the average food truck,” Casar said in a statement. “That must change.”

A summary of the legislation did take a stab at defining the elusive term, offering two possible thresholds for superintelligence. A system qualifies if it:

• “exhibits or can easily be modified to exhibit capabilities that match or exceed human cognitive performance and capabilities across a broad range of domains or tasks”; or
• has “sufficient capabilities to plan and execute the disempowerment of humanity, including by overthrowing or undermining the U.S. government.”

By the first of those measures, the ban may already be too late, some researchers say. “I don’t agree with the definition, but I believe it has already been met,” says François Chollet, who created a test—called the Abstraction and Reasoning Corpus for Artificial General Intelligence—for measuring machine intelligence. For example, Chollet thinks some frontier large language models already exceed human cognitive capabilities in some areas.

Others argue the definition is too loose to bind anyone. “The inconsistent and unfalsifiable nature of these definitions will likely exempt AI companies from this legislation,” Khlaaf says. “They can claim plausible deniability, as the noted definitions are either ill-defined, hypothetical, or focus on risks that do not currently exist.”

Sanders’s office said further details about how regulators would identify superintelligence would come when the bill is formally introduced in the coming weeks. But an outline released by the lawmakers said it will call for establishing a new cabinet-level federal agency, advised by a panel of independent AI experts, to enforce the prohibition. The bill would also pause advanced AI development by companies until that agency sets safety standards. Firms or institutions that violate the law would face the “corporate death penalty,” or forced dissolution, and individuals could face prison terms of up to 20 years. The bill also aims to lay the groundwork to ban superintelligence worldwide by encouraging the U.S. government to pursue international agreements.

Debates over AI terminology are not new. Many use terms such as artificial general intelligence (AGI) and superintelligence interchangeably, whereas others argue the two are separate.

The major AI companies are equally divided. OpenAI’s charter, for example, defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” In a briefing on Thursday, company President Greg Brockman proclaimed its new Astra model exhibits AGI. “If we fast forward a couple years, and we look back and say when was it really that AGI was created … I think it might be about this model,” he said. “I think it’s not unreasonable to feel that we are now in the AGI era.”

Anthropic CEO Dario Amodei, meanwhile, wrote in his 2024 essay “Machines of Loving Grace” that he finds “AGI to be an imprecise term that has gathered a lot of sci-fi baggage and hype.” Instead, he pushes for the term “powerful AI,” which would be equivalent to a “country of geniuses in a data center.”

For some, regulation of AI should not wait for a linguistic consensus. Computer scientist Roman Yampolskiy of the University of Louisville calls the proposed legislation “a step in the right direction,” and says uncertain definitions do not make regulation impossible. “Laws routinely regulate categories whose boundaries require expert interpretation. More importantly, regulations should not wait until we can prove that a system is already superintelligent. By that point it may be too late.”

With elections looming in November, Congress is unlikely to act on the proposed legislation this year. But with U.S. voters becoming increasingly uneasy with the rapid advancement of AI, it is likely to be joined by other proposed laws requiring the government to step in.

LISTEN TO THIS ARTICLE

It sounds like an orca rampage. Over the past 6 years, killer whales have been ramming boats off the coasts of Spain and Portugal, sinking nine, and parading fiberglass remnants around like trophies. Vessels are frequently towed into shipyards with tooth rakes on their undersides and rudders ripped off, causing millions of dollars in damage.

Sailors have had enough. They’ve hurled homemade explosives at the critically endangered animals and even fired shotguns at them . The boaters are fending off what they believe are deliberate assaults by killer whales waging a vendetta after being struck by yachts and getting tangled in fishing gear. But are the giant cetaceans really out for blood?

Not according to a recent study in Marine Mammal Science . Based on underwater footage and firsthand observations, scientists conclude the killer whales aren’t trying to harm humans . Instead, they probably just want to play.

Theories abound over why Iberian orcas ( Orcinus orca ) began to batter boats. The most sensational holds that the whales are exacting revenge for injuries inflicted by humans. Orcas can become entangled in nylon fishing lines as they try to pry away hooked bluefin tuna, sometimes gashing their sides or amputating a fin. Recreational boats also soar through their waters, risking collisions. Popular culture holds that a supposed matriarch, White Gladis, was hit by a yacht; traumatized, she began to teach younger whales to dish out payback .

Not all researchers are on board. “This notion of retaliatory behavior is absolutely ridiculous,” says Jared Towers, a killer whale scientist at Bay Cetology who wasn’t involved in the current study. “Wild orcas that get harmed don’t attack; they swim away.”

Other scientists suspect the animals are training their young to hunt, using the boats to practice the coordination and motor skills they’ll need for capturing prey. But the orcas’ boat interactions show very different tactics compared with hunting, which involves behaviors such as high-speed chases and erratic underwater movements, notes Janek Andre, an orca researcher at WeWhale who did not participate in the new work.

Joana Castro, a marine ecologist at the Marine Environment Research Association (AIMM), has found a more benign explanation. Over the past 8 years, she and her colleagues have analyzed seven encounters they’ve filmed during routine surveys of large sea animals along the southern coast of Portugal in their single-engine, 6.5-meter inflatable boat. Orca interactions with the vessel ranged from light touches to full-on high-speed ramming that knocked some students—who’d tagged along on the expeditions—to the floor, says Guilherme Estrela, a skipper and biologist at AIMM. To get a picture of what was happening underwater, brave participants plunged hand-held GoPros below the surface.

In a few videos, the orcas could be seen opening their mouths or blowing bubbles—behaviors scientists often associate with aggression, Castro says. But the rest of the evidence seemed to fit more with the idea that the animals were simply playing: Almost all the actual boat contact was made by juveniles—“teenagers who do reckless things”—and never adult males, Rose says. And the animals’ contact was often simple touching or gentle nudging, Castro adds. Sometimes, they would show their bellies—a friendly posture—while pushing the boats.

Some of the orcas also came bearing gifts. In 2021, for example, 3-year-old Eli dragged a chunk of tuna in her mouth to show the researchers. “That’s supersocial and bonding in their own system,” Castro says. “A killer whale does not do that to an enemy.”

Andre has had similar experiences with Estrella —the one boaters call White Gladis—and her 5-year-old son Atlas, who bears deep scars from a 2022 propeller strike. “If you want to avenge yourself, you don’t bring food, you know?” he says.

Play usually gets its start in exploratory behavior, and Castro and her colleagues appear to have caught those early origins as well. In a brief clip from 2018, 2 years before the ramming incidents started, 6-year-old Lua nudges the boat’s inflatable pontoons as his little sister Sintra watches intensely. “At first, we didn’t make a lot of it,” Castro says. “But then, well, it escalated.” Both Lua and Sintra would later become confirmed boat rammers.

Regardless of the killer whales’ true intentions, the behavior has concerned scientists and conservationists, especially because Iberian orcas—a genetically distinct subgroup of about 40 individuals—are critically endangered. The animals are learning to play a game that could get them killed, Castro’s team says, either by angry boaters or by propellers and other dangerous boat equipment.

Experts offer different solutions, from boats speeding up so they won’t be rammed to stopping dead in their tracks when a whale approaches. But Andre says the best strategy is to avoid the orcas altogether. “Don’t engage them; don’t call their attention,” she says. “Just stay away.”

LISTEN TO THIS ARTICLE

When President Donald Trump’s administration announced in August that it was proposing to rescind a 25-year-old U.S. Forest Service (USFS) regulation barring road construction on some national forest land, officials billed it as a way to reduce wildfire dangers and aid overgrown forests. “Outdated restrictions have kept tens of millions of forested acres off-limits to the very treatments that improve forest health and reduce wildfire risk to our communities,” Department of Agriculture Secretary Brooke Rollins stated in a press release.

But research, including work by USFS’s own scientists, suggests the road-building restrictions haven’t exacerbated wildfire risk in recent decades. Although some experts think removing the restriction will allow needed forest thinning, others say it could damage ecosystems and fuel more wildfires.

“It’s infuriating that the potential for reducing fire risk is even used as a justification for putting more roads into our national forests,” says Jennifer Balch, a University of Colorado Boulder fire scientist. Balch’s work has shown that humans start 84% of all wildfires nationwide, and that roads are a major way people carry matches, fireworks, and other spark-producing tools into the woods.

The roadless area conservation rule, better known as the roadless rule, was issued in 2001 in the waning days of then-President Bill Clinton’s administration. The regulation banned road construction on more than 175,000 square kilometers of roadless areas, nearly one-quarter of all national forest land. Since then, it has been embroiled in a near-constant legal and political tug of war.

In 2020, the first Trump administration rescinded the rule in Alaska’s Tongass National Forest, opening it to loggers. Former President Joseph Biden’s administration reversed that change in 2021. Now, the second Trump administration is going a step further and proposing to eliminate the rule nationwide. Public comments on the proposed change will be accepted through 21 September .

In a draft environmental impact statement , which spells out the science underpinning the proposed repeal of the rule, USFS notes that more than 29,000 square kilometers of land covered by the roadless rule lie within forests designated as high priority “firesheds”—places posing an elevated wildfire risk to infrastructure and communities. Roughly half that area might benefit from logging or thinning to reduce fire risks, the analysis found.

The current roadless rule hampers those efforts, says Gordy Sanders—a longtime forester and chair of the Montana Forest Collaboration Network, an organization that promotes forest management collaborations between different interest groups. He says that in his experience, USFS officials are often leery of pursuing forest treatments that involve even a sliver of designated roadless forest, because of the potential for legal and paperwork headaches. “There’s a lot of projects that didn’t move forward but really should have,” Sanders says.

But others question whether the constraints have raised wildfire risk in roadless areas. In 2020, USFS ecologist Sean Healey published a paper in Environmental Research Letters reporting that between 1985 and 2016, national forest acreage burned at approximately the same rate regardless of whether it was shielded from road construction. A study released earlier this year as a preprint also found land covered by the roadless rule had moderate or severe fires at roughly the same rate as national forest land open to road construction. “We don’t see any evidence that the roadless areas are burning more. It just hasn’t happened that way,” says study co-author and Oregon State University geographer Robert Kennedy.

Alexandra Syphard, a wildfire scientist and senior researcher at the Oregon-based nonprofit Conservation Biology Institute, began circulating a letter last month critiquing USFS’s environmental impact analysis. She says some 30 prominent wildfire scientists added their names to the letter within days. “The notion that the rescission of the rule will lead to improvements in wildfire risk reduction is not evidenced by the science,” Syphard says.

Scientists have also voiced concerns about the detrimental ecosystem impacts of roads, pointing to evidence that they are a conduit for invasive plants such as grasses—which raise fire risks—and that their construction can unleash more sediment pollution into nearby creeks and rivers. Areas subject to the roadless rule contain more than 130,000 kilometers of river and help protect watersheds supplying water to at least 25 million people, scientists reported this year in PLOS Water .

James Johnston, a University of Oregon fire scientist who has worked for years on a collaborative forest management program in eastern Oregon’s Malheur National Forest, says he’d like to see the roadless limits applied more flexibly. He’s watched USFS drop several worthwhile projects amid fears of legal problems. “Environmental organizations have been eager to challenge any activity in roadless areas in court,” he says.

But he thinks killing the rule is too draconian. “The vast majority of … roadless areas are not the problem,” he says. “Simply rescinding the 2001 rule is almost certainly not the solution.”

LISTEN TO THIS ARTICLE

The worsening water crisis in the Colorado River Basin stems from a century of overuse and, more recently, a sustained drought. Short-sighted water allocation policies are responsible for the overuse, scientists say, and human-induced climate change is a major contributor to the drought.

The National Science Foundation (NSF) is betting technology can relieve the crisis by better managing the water that remains and tapping new sources. The Energy Water Security Initiative (EWSI), now open for proposals, asks “multisector” research teams to propose technical fixes designed to ensure that the Colorado can continue to support both the economies of the seven states that rely on the water and the 60 million people living there.

But many scientists who have spent their careers studying the basin say the initiative, launched with a $50 million down payment on an investment that could eventually exceed $100 million, is misguided. “These guys have completely missed the point,” says Jack Schmidt, a river scientist at Utah State University. “This initiative has nothing to do with addressing the actual crisis, which is the dwindling flow of water in the river.”

Others are withholding judgment. “If NSF does it right, it could result in some real benefits for the region,” says paleoclimatologist Jonathan Overpeck, dean of the University of Michigan’s School for Environment and Sustainability.

Like many recent technology-focused initiatives at NSF, the EWSI will short-circuit the agency’s traditional merit-review process and divert funding from basic research, including studies of the climate shifts that have worsened the crisis. To University of Arizona (U of A) geologist Karl Flessa, a former NSF program officer, the EWSI is the latest example of how the White House is interfering with NSF operations and channeling its $8.8 billion budget toward the administration’s research priorities. “Where can the White House find cash that it can move quickly and without much merit review?” Flessa asks. “The NSF piggy bank.”

NSF declined comment on the initiative, which was announced on 23 July.

In the past, the agency has addressed water crises like the Colorado’s by supporting climate and hydrology studies and research designed to help communities better cope with the interaction of climate change, population growth, and resource depletion. A recent NSF program called the Sustainable Regional Systems Research Networks, for example, gave up to $15 million over 5 years to urban-rural collaborations trying to provide sufficient energy, water, food, and health services to communities in crisis. But that program was terminated after President Donald Trump took office, along with many other NSF activities Trump deemed “woke” or insufficiently rigorous to warrant the use of tax dollars.

The EWSI has a different focus: maintaining adequate water levels for agriculture, by far the largest user of the river’s water, as well as for energy generation and the growing number of water-hungry semiconductor fabrication plants and data centers. The fixes, according to the solicitation, might include “AI-informed hydropower modeling, reservoir evaporation suppression, closed-loop recycled-water cooling for data centers, brackish groundwater desalination, and critical mineral recovery from wastewater.”

Researchers don’t quibble with the EWSI’s goal of helping “secure American energy and national infrastructure resilience by producing measurable reductions in vulnerabilities [caused] by water scarcity,” according to the solicitation. But many take issue with its assumption that “accelerating deployment of technical solutions” is sufficient. “It really isn’t a technology problem. It’s a governance problem,” says Andrea Gerlak, a political scientist at U of A who studies how water resources are managed. “We gave away too much water, and we need to address that.”

NSF is also breaking with tradition in how it will review EWSI proposals. Instead of asking outside experts to vet projects based on both their intellectual merit and broader benefit to society, the agency is using a funding mechanism, called a Broad Agency Announcement (BAA), that doesn’t include the “broader impacts” criterion and doesn’t specify who will review the proposals.

The BAA imposes an extremely tight deadline on applicants. Five-page white papers describing the proposed work were due on 10 August, only 18 days after NSF unveiled the EWSI. Those who make it past an initial screening have 6 weeks to submit a full proposal.

The short timeline and streamlined reviews raise questions about potential bias in the selection process, Overpeck says. “I’m not saying that EWSI is a bad thing, but the peer review better be good and the results of the work better be tangible and worth the money,” he says.

NSF is financing the EWSI through its geosciences directorate, whose pot of money to support fundamental research in its core areas of earth, atmospheric, and ocean sciences has been trimmed by one-third or more under Trump. “If we lived in a world of infinite money, none of this would potentially matter,” Schmidt says. “But resources are finite. And when it comes to supporting university research, it’s really a world of declining resources.”

That reality, he says, prompted his university—and Arizona, Michigan, and most likely dozens of other research institutions—to respond to the solicitation despite the tight deadline. “A few hours after that thing was posted, an email went out saying there is money on the table,” Schmidt recalls. “The next day there was a Zoom meeting and the message was full steam ahead.”

“Universities are so desperate for money,” he adds, “that they’re going to chase whatever is available.”

When Nature Medicine published a paper in January 2025 showing users of popular weight loss drugs have a reduced risk of a wide range of disorders, the reputable Dutch newspaper de Volkskrant was one of many media outlets around the world to cover the news. Its story, headlined “Ozempic proves surprisingly effective against 42 conditions,” said glucagonlike peptide-1(GLP-1) receptor agonists “not only make many kilos disappear. They also have surprisingly beneficial health effects.” The story explained the study was observational and that 19 different side effects had been reported for the drugs. The paper’s editors felt it was a balanced piece.

But Dutch law enforcement officials disagree. Last month, the Dutch Health and Youth Care Inspectorate (IGJ) fined publishing company DPG Media €51,000 for the story in de Volkskrant and another in the AD newspaper, alleging they violated an EU-wide prohibition of promoting prescription medicines to the public. IGJ also imposed hefty fines on several other prominent Dutch newspapers, a lifestyle magazine, and a popular talk show for their coverage of GLP-1 receptor agonists.

Media in at least two other countries have run afoul of Europe’s strict rules around pharmaceutical advertising as well, in most cases because they wrote too enthusiastically about the new generation of weight loss drugs.

Editors-in-chief in the Netherlands, along with the Dutch journalists’ union, have voiced outrage at IGJ’s decision, warning of censorship and a “chilling effect” on journalism. “The article for which we are being fined meets all our journalistic criteria and protocols,” Volkskrant Editor-in-Chief Pieter Klok writes in an email to Science . “If this piece is not allowed, we can never write about medicines again.”

“We describe and analyze reality. In doing so, we are not promoting it,” adds science journalist Lucas Brouwers, deputy editor-in-chief of the NRC newspaper, which was fined for two similar stories.

But tone matters under the law, says Gertie Lintjens, a life sciences lawyer at international law firm CMS. “Any communication about a prescription-only drug that gives the impression you are promoting this product is simply not allowed,” she says.

Roland Seifert, a pharmacologist at Hannover Medical School who studies how news media cover pharmaceuticals, agrees with IGJ that some of the Dutch stories—including the one in de Volkskrant —are “overly positive” and “may raise unrealistic expectations in the public.”

In contrast to the United States, where TV viewers are bombarded with drug ads, a 2001 EU directive prohibits all activities “designed to promote the prescription, supply, sale or consumption of medicinal products” to the general public, with the exception of vaccines that are part of official vaccination campaigns. The goal is to make sure doctors, not advertising budgets, determine which medicines are prescribed, prevent medical overconsumption, and help keep health costs in check. The law applies not only to pharmaceutical companies, but to any person or group targeting the general public, including patient organizations, health influencers, and media.

A 2009 ruling by the Court of Justice of the European Union established key case law. In a case against a Danish journalist who posted information on his website about a rosehip powder, the court ruled that even “a message originating from an independent third party may constitute advertising,” and that a message does not have to be “disseminated in the context of commercial or industrial activity” to count as promotion.

“You can promote a product in many different ways,” Lintjens says. Problematic elements include repeatedly using a drug’s brand name instead of its generic name, superlatives such as “miracle cure,” and an overly positive tone. “It is often just one or two sentences that push the communication over the edge,” Lintjens says. But she acknowledges the line between information and promotion can be blurry.

DPG Media has said it will appeal the fines in court, but it has lost two similar cases in the past, one involving an article published in the regional newspaper De Stentor in 2023—headlined “Jeroen weighed 186 kilos at his lowest point, now he uses a new diet medication that is not yet on the market”—and the other a cover article in magazine Flair , also in 2023.

Journalists in Switzerland—which is not an EU member but keeps its laws in line with EU directives on many issues—have fought back against similar verdicts. The Swiss Agency for Therapeutic Products has ordered several articles removed from the website of the newspaper Neue Zürcher Zeitung , some of them about GLP-1 receptor agonists. The newspaper lost an appeal in one case; two other appeals are still pending. U.K. authorities have clamped down on coverage of GLP-1 products as well.

Part of the Dutch clash may stem from a shift toward more popular, accessible science journalism over the past 2 decades, Lintjens says, with more stories reflecting the emotions of patients, doctors, and researchers. “Such a tone of voice is harder to reconcile with the neutral tone the law demands,” Lintjens says. But rather than simply invoke freedom of expression in their court defenses, she thinks media should explain how their stories are intended to inform the public or fuel debate: “That is where the legal space is.”

Chemotherapy has helped transform many cancers from fatal diseases into manageable or treatable conditions. It also causes myriad side effects, including nerve damage and sometimes lasting pain, severely affecting people’s quality of life long after they recover from cancer.

But a compound found in psychedelic mushrooms may help, researchers say. Giving mice psilocybin before chemotherapy protected them against nerve damage and pain without interfering with the treatment itself, according to a study published today in Science . And lab experiments suggest the drug blocks chemotherapy-associated cellular and molecular changes in mouse and human nerve cells. An upcoming clinical trial will test the compound in people undergoing cancer treatment.

If confirmed, the findings could have huge implications, says Alex Kwan, a neuroscientist at Cornell University who was not involved in the work. Psilocybin’s effects in this study are “quite incredible,” though he and other researchers say they’d like to see more data to be convinced of the drug’s potential in people.

Chemotherapy works by targeting fast-growing cancer cells, but often also hits healthy cells in the body. Platinum-based drugs such as cisplatin and other widely used therapies can damage peripheral nerves, causing tingling, burning, or numbness, especially in the hands and feet. There are no approved treatments for these side effects; doctors sometimes prescribe duloxetine, a drug approved for depression and other disorders, but research suggests it fails to alleviate chemo pain in many people.

Psilocybin, found in various “magic” mushroom species, has received increasing attention in recent years as a possible pain medication, as well as a treatment for psychiatric disorders. The human body breaks it down into a compound called psilocin, which interacts with receptors for the neurotransmitter serotonin, altering brain activity and contributing to hallucinations and other reactions.

Evidence for psilocybin’s effectiveness as a painkiller is mixed. But researchers at the MD Anderson Cancer Center wondered whether giving the drug in advance, rather than after nerve damage has occurred, might be more fruitful. “The idea, instead of trying to put things back together that are broken, is to jump in and protect it from the very beginning,” says study co-author Patrick Dougherty, a pain researcher at MD Anderson.

As expected, the team found mice given chemotherapy showed peripheral nerve damage and were hypersensitive to pain for at least 4 months after treatment. But giving mice a dose of psilocybin before chemotherapy shortened that hypersensitivity to about 3 months. Two doses given 1 week apart eliminated it altogether.

Importantly, psilocybin didn’t influence tumor growth or chemotherapy’s effectiveness in cancerous mice, the authors stress. And although most of the team’s experiments used cisplatin, tests with other, nonplatinum chemotherapy drugs produced similar findings, suggesting psilocybin’s benefits aren’t limited by the type of cancer treatment.

To understand how psilocybin was preventing pain, the team studied mouse nerve cells in live animals and in a dish. They found the compound blocked some of the harmful cellular and molecular effects of cisplatin. For example, cisplatin disrupts the transport of organelles called mitochondria—the cell’s power plants—along neurons, but advance treatment with psilocybin prevented this in mouse cells. Further experiments suggested the compound similarly protected human neurons, too.

Some of the team’s analyses of these mechanisms remain “preliminary,” says Giampietro Schiavo, a cellular neurobiologist at University College London who was not involved in the work. It will be important to understand how transport of organelles other than mitochondria are affected, as these could also be relevant to how psilocybin works, he adds. Nevertheless, “Finding [treatments for chemo pain] that do not affect the efficiency of the chemotherapy is absolutely crucial,” Schiavo says, and “any advance in this area is phenomenal.”

Susanna Park, a neuroscientist at the University of Sydney, urges caution in interpreting the team’s findings for patients. As in many fields, “there has been a large gap between success in preclinical models [of chemotherapy pain] and translation to effective therapies for humans,” she adds. She’d like to see more data on psilocybin’s impact on larger nerve fibers, which are particularly damaged by chemotherapy, to confirm a truly neuroprotective effect.

MD Anderson plans to launch a phase 2 trial next month to evaluate psilocybin in about 80 patients receiving chemotherapy for breast, colorectal, or head and neck cancer. MD Anderson surgeon-scientist and study co-author Moran Amit says he expects the trial to fill up fast. But it will be years before the team has collected sufficient data on the drug’s long-term safety and efficacy.

The hallucinations and other sensations psilocybin causes may be difficult for some patients to tolerate, Kwan notes. But in additional tests, the researchers showed an experimental synthetic, psilocybinlike compound that wouldn’t make patients trip replicated some of the protective effects of psilocybin in mice.

Altogether, the multiple findings in this study “open up some new ideas that have not been thought about in this field,” Kwan says. “I’ll be very excited to see where it leads.”

You’re an editor or a peer reviewer, and unread manuscripts are piling up on your desk. Some of your colleagues are trying out reviewing tools powered by artificial intelligence (AI) that could help relieve the burden. Some tools check for incorrect citations or calculations. Others promise a judgment on whether the manuscript’s claims hold up, or even whether they advance knowledge in that field. Would you read those assessments? Trust them? Criticize or reject a paper based on them?

Those questions are looming large for a scientific community that has spent years struggling with an overburdened peer-review system and now faces immense new challenges created by generative AI. The number of scientific publications has doubled in 5 years, to more than 8 million in 2025 ( see graphic, below ), and some publishers report an even steeper rise in submissions—fueled in part by the wider availability of large language models (LLMs), which can churn out some drafts in minutes. Editors are increasingly struggling to find willing reviewers, contributing to lag times between submission and publication that can drag on for months to more than 1 year. To get just one completed review, editors needed to send an average of 4.5 invitations in 2025—double the number required in 2018 , according to a report this year by the Silverchair content-hosting company.

Some researchers and publishers think AI will have to be part of the solution. Automating some of the peer-review process could help speed publication and save researchers’ time, advocates say. And, in fact, a growing number of reviewers are employing these tools: In a global survey of 1645 researchers across multiple scientific fields by the publisher Frontiers, just over half of respondents said they used AI for review tasks in 2025 , including to help draft comments and in some cases to review methods and statistics. That’s up from 29% the previous year—despite many journals prohibiting those uses.

But others are skeptical that machines can give reliable feedback, especially for the most important, core aspects of peer review, including assessing novelty and significance, while avoiding risks such as bias and gaming. “The peer-review crisis is real, and AI looks promising [to help], but irresponsibly rushing towards automating a lot of the judgment might backfire,” says Joachim Baumann, a postdoctoral researcher at Stanford University who studies the societal impact of AI. “We need to make sure the tools we deploy are fit for the task.”

Research projects and pilots testing AI’s capabilities have surged in recent years, aimed at doing just that. “If we are able eventually to get large language models to really do this properly—and increasingly they’re obviously getting better—I definitely think they could be a way out of this huge quagmire,” says Tom Hope, a researcher at the Allen Institute for AI and the Hebrew University of Jerusalem who is developing AI review tools. As Stephen Turner, a data scientist at the University of Virginia (UVA), sees it, “The question is not whether AI matches the best human review. It is whether AI-assisted review with clear rubrics outperforms the inconsistent, fatigue-influenced, mood-dependent median review.”

For now, most journals are largely barring reviewers from using AI tools until editors determine how they can best support high-quality reviews. Often, however, “People are using these tools in a clandestine, unregulated way,” Turner says. “It’s kind of the Wild West out there now.”

With that in mind, Science offers this guide to what researchers are discovering about AI-assisted review and how it might be used productively.

Human reviews have quirks

Evaluations by human reviewers are a vital ingredient in scholarly communication. However, they can be frustratingly subjective and superficial, says Agnieszka Swiatecka-Urban, a physician and researcher in pediatric nephrology at UVA. When humans have flagged deficiencies in her manuscripts, some have written little more than that they “diminish their enthusiasm” for the results. “It’s quite irritating because it does not focus on the science,” she says. “And there is a temptation—what I see from many reviewers—just to skim through and not be thorough.” Human reviewers have also been shown to favor work by prominent researchers and to make harsh, critical comments.

Swiatecka-Urban sees reasons why AI review, if properly designed, can complement or surpass human reviewers. She recently asked AI for suggestions on how to improve several of her manuscripts and a grant application, and found the results useful—for example, in assessing whether the conclusions followed from the results and the same numbers were consistently reported across text, tables, and figures. “What impressed me the most was the objectivity,” she says.

Studies bear out that machine-generated reviews may also be more consistent: When both AIs and people were asked to rate conference manuscripts, the machine ratings clustered more tightly than human reviewers’ , according to a recent study by Baumann and colleagues. Some see this as positive, a correction to inattentive human reviewers. But Baumann warns such clustering could also promote an “intellectual monoculture” that penalizes unorthodox ideas.

Other scholars value the variety of feedback provided by traditional peer review. “What I like about human reviews is that you’ll get three different reviews and they’re all three completely different—people always pick up on different things,” says Ruth Ley, a microbiologist at the Max Planck Institute for Biology. “It just seems sad to lose the heterogeneity of the process.”

Reviewers often diverge because of the inherent subjectivity of assessing whether a manuscript’s findings are novel—often cited as a key responsibility of peer review, but notoriously thorny to define and pin down. Hope and colleagues are among the teams working to develop AI’s capacity to help. Their method uses an LLM to extract key claims about novelty from a manuscript; search online for and summarize related, previously published papers; and evaluate how the new paper departs from that record. In a study of 182 submissions to a recent machine learning conference, human evaluators judged the AI reviews as at least as good at evaluating novelty as human-written ones in most cases . But the approach may miss paradigm-shifting contributions that human experts would recognize and may struggle to keep up with fast-moving fields, Hope acknowledges. Assessing novelty is “a very interesting problem because it’s so hard to evaluate,” he says.

What AI peer review does well

One key area in which LLMs may shine is performing routinized, tedious checks for correctness that time-pressed human reviewers often skip. For years, publishers have employed automated checks to make sure manuscripts aren’t plagiarized, transparently report methods, and meet other standards. LLMs could bring an expansion of those capabilities, recent research suggests.

In one study, researchers had OpenAI’s GPT-5 evaluate 2500 randomly selected papers accepted by three leading machine learning conferences from 2018 to ’25. They found the AI was successful at assessing each paper on specific, narrow criteria, such as whether the calculations were correct and the text matched facts in tables, researchers reported in a December 2025 arXiv preprint. The AI found that about one-third of the papers contained at least one error that could affect interpretation, reproducibility, or downstream use—and human evaluators confirmed 83% of those errors, although they found most did not change the paper’s main conclusions. The AI was also better at detecting math errors than text mistakes. “Human reviewers may not have the time to go through [a paper] line by line,” says study co-author James Zou, an AI researcher at Stanford. “That speaks to a complementarity where AI can help.”

Another recent study, reported in May in a preprint on arXiv, undertook a granular, head-to-head comparison of human and AI critiques of the same manuscript. Overall, it found AI reviewers were more diligent than human reviewers at routine but labor-intensive tasks, such as autonomously executing computer code submitted with the manuscript to check whether it supported the paper’s claims.

In that study—one of the most extensive of its kind—Ph.D. student Seungone Kim of Carnegie Mellon University and colleagues gathered the preprint versions of 82 papers published in journals in the Nature family as well as human-written reviews of those manuscripts, which some of the journals have begun to post with the final published version. They directed three popular LLMs—versions of OpenAI’s GPT, Anthropic’s Claude, and Google’s Gemini—to provide structured reviews of the preprints. The team then asked 45 scientists to score whether each of the 3000 individual critiques in the AI reviews was correct, evidence-based, and significant—meaning it could meaningfully improve the paper—and to identify the best human review for each paper.

Across all papers, GPT scored better than the top-rated human on a composite of the three criteria, and all three LLM models scored better than the lowest rated human . About one-quarter of the AI criticisms were not picked up by any human reviewer and were not trivial, the authors reported. (The AIs also missed the mark in some key areas— see “Where AI reviews fall short” .)

Some authors are beginning to have access to checkers designed for scientific manuscripts and report finding them useful for improving their drafts prior to submission. This year, three computer science meetings offered authors the opportunity to use Google’s Gemini-based Paper Assistant Tool, which inspects methodologies, calculations, and proofs. Among the 1068 authors who used it to prepare for the Conference on Neural Information Processing Systems (NeurIPS), 86% said they found it very or mostly helpful, Google researchers say.

To address concerns that parts of unpublished manuscripts fed through AIs could be used to train an LLM or otherwise become public, the Paper Assistant Tool and other manuscript evaluation services promise to keep the uploaded manuscript in a private repository unconnected to the internet. NeurIPS says it did not show the AI reviews to the conference’s referees.

In recent months, the bioRxiv preprint server also began to offer authors the option of sending manuscripts to one of several AI technical-review checkers. The reviews remain confidential to the authors, who can decide whether to revise the manuscript before it appears. Some users report positive experiences, for example with one checker called q.e.d. Science. “I had a more rigorous, attentive, constructive, and useful experience with q.e.d. than with many of my peer reviewers in my 35-plus years submitting papers,” says Michael Levin, a developmental biologist and bioengineer at Tufts University who used the service last year.

Where AI reviews fall short

Although Kim’s study found AI review has strengths, it also highlighted shortcomings. The LLMs, for example, scored less well overall than human reviewers on the correctness of their assessments. One faulted a paper for not correcting for a known bias in data about air pollution in China—even though the study actually had. Kim’s team attributed the mistake to a known limitation of LLMs: Constraints on their working memory may cause them to overlook relevant details, especially when these are spread across multiple sections of a paper and supplementary materials.

The LLMs also displayed a limited grasp of methodological conventions used in certain subdisciplines. For example, a different AI reviewer criticized a particle physics paper produced by a team using the Large Hadron Collider because it did not provide enough detail on the data analysis it used, limiting the study’s reproducibility. But such details are routinely housed centrally at the collider’s parent lab, CERN, a convention of particle physics the AI did not understand, a human reviewer noted.

In sum, Kim and colleagues write, “AI reviewers could complement, but should not replace, human reviewers. … AI reviewers and human reviewers converge on which parts of a paper warrant review but diverge in how they characterize what they find, meaning that an AI panel is not a drop-in replacement for a human panel.”

AI reviews may also make unreasonable requests for additional results. When Ley ran one of her published manuscripts through q.e.d. Science, the AI suggested adding details that would have required time-consuming follow-up studies not necessary to establish the paper’s main point, she says. “It brought up things that we know would be the next step, the next paper,” she says—but they “would take a ridiculous amount of time. … That’s where human judgment would come in.”

Some scientists add that AI reviews are verbose and tend to nitpick minor problems while failing to highlight the highest priority issues. Yoshitomo Matsubara, a researcher at the technology company Yahoo, described an AI-written review he saw as part of a pilot for this year’s annual meeting of the Association for the Advancement of Artificial Intelligence (AAAI) that tested using AIs to evaluate conference submissions ( see “Conferences and journals are testing the waters” ). The review was separated into three sections containing 34 bullet points, totaling 14,000 characters; his own review of the same paper totaled just 4000. “The AI review is too lengthy, and these dots are not well connected,” Matsubara says. “It’s too much information.” And he deemed a summary generated by the AI to be superficial.

The machines can also reflect baked-in biases from the vast body of text on which the LLMs were trained, such as a preference for work by authors from high-prestige institutions. In one experiment, researchers asked an LLM to evaluate multiple versions of papers published in the Proceedings of the National Academy of Sciences , identical except for the lead author’s name and affiliation, which the team changed. The LLM was more likely to recommend rejection for papers by authors at lower prestige institutions than those at prominent ones, the group led by computational social scientist Anthony Howell of Arizona State University reported in a 2025 arXiv preprint.

Furthermore, AI reviewers struggle to assess results in new fields or narrow subdisciplines in which few prior studies are available to the LLM to inform its judgments. Reflecting AI’s tendency toward sycophancy, several studies have found that the machines tend to award scores to manuscripts that are, on average, higher than those from human reviewers. And Baumann’s research has shown AI reviewers can be gamed into giving a paper a higher score if authors explicitly direct an LLM to rewrite the paper for that purpose, a trick he calls paper laundering.

At the end of the day, using AI may not even save reviewers time . A research group asked 24 scientists to review two similar published papers selected by the group; the reviewers were directed to use AI assistance for only one of the two reviews. When using AI, the reviewers spent less time on tasks such as checking other relevant literature, but that was offset by the need to check the accuracy of the AI’s suggestions, according to a conference presentation last year by a team from University College London (UCL). The net time saved, a few minutes on an hourlong review, was not statistically significant.

Given the limitations of AI, humans need to resist its allure, says Duncan Brumby, a member of that study’s research team who studies human-computer interactions. “The more fluent and effortless the [AI] output becomes, the easier it may be to accept something that sounds like a considered review without having done the corresponding intellectual work.”

Conferences and journals are testing the waters

Even amid the uncertainty surrounding AI reviews, and a lack of standards for using them , some journal editors and conferences are dipping their toes into real-world experimentation. When Joydeep Biswas, a researcher in robotics at the University of Texas at Austin, and colleagues were organizing this year’s AAAI meeting, they were worried about the rapid growth of submissions and the burden on reviewers. Given the conference’s topic, they felt a responsibility and opportunity to study this “socio-technical problem,” and whether AI could help ease the load, Biswas says. Still, getting agreement to do the pilot wasn’t easy. “I’ve been on the executive council for a few years and I’ve never had a discussion which was as contentious and as time-consuming,” he says.

They decided their goal would be to test and gather data for how well AI could do the job, while still including ample human input and limiting the extent to which the AI reviews would influence the ultimate decision to accept or reject each submission. They applied a team of LLMs, working together, to help process the 30,000 submissions received, double the number for the previous year’s conference. Authors received one AI-written review in addition to at least two human reviews. The human reviews were the only ones that included numerical scores and recommendations about whether the submission should be accepted, and were written without seeing the AI evaluation. Authors then drafted rebuttals to inform final decisions by the conference’s human judges.

In a survey of participants, a majority of nearly 6000 authors, reviewers, and judges said they preferred the AI-written reviews over the human ones on six of nine dimensions of quality , such as whether a review was thorough and suggested useful improvements to the research design. Still, authors preferred human reviews over AI in some important categories, including whether the review contained technical errors and overemphasized minor issues. Authors tended to view the AI reviews more favorably than the reviewers and judges. A single AI review “cannot satisfy all three audiences,” Biswas says. Authors are eager to improve their paper, whereas judges need a top-line synthesis about whether it makes an original contribution to the field—which often requires intuition only a human can provide. “I came away fully appreciating the importance of human expertise,” he says.

“I think there is a way of using humans plus tools so that a human can address a larger number of papers, make better use of their time, and every paper gets some amount of human attention,” Biswas adds. He says AAAI plans to repeat the pilot for its 2027 meeting but has no plans to reject papers based solely on AI reviews—a move he expects would draw “stiff opposition” from attendees.

Organizers of another meeting on machine learning, the 2025 International Conference on Learning Representations, ran a trial to see whether LLMs could improve reviews, based on the LLM’s analysis of the submitted manuscript and a human-drafted review. After receiving the AI’s feedback, 27% of the human reviewers updated their critiques , making them longer on average, and almost all included at least one point from the AI suggestions. A separate panel of human evaluators judged that two-thirds of these revised, AI-informed reviews were better than the original. “We were pleasantly surprised at how many reviewers incorporated the AI feedback, because we know how busy reviewers are,” says Zou, senior author of a February paper in Nature Machine Intelligence that describes the experiment.

But when the UCL team interviewed reviewers at the same conference, they found mixed views. Some considered the AI critiques to be redundant , or even patronizing because they suggested adding excessive detail, according to the study. “I’m probably confident in my review rather than believing something else [the AI],” one of them told the researchers.

On the journal side, NEJM AI , an affiliate of The New England Journal of Medicine focused on the use of AI in medicine, began a pilot last year offering a “fast track”: an accept-or-reject decision within 7 days based on a combination of AI analysis and review by its human editors. The track is only offered for papers editors think would have a high chance of acceptance through the conventional route. In lieu of sending the paper for outside peer review, an editor performs a full review without AI’s help, and other editors debate the human and AI reviews before deciding whether to publish the paper.

Since the journal published the first two papers through this track in November 2025, the quality of the AI analysis has been improving, and so have authors’ and editors’ comfort levels, says Senior Deputy Editor Arjun Manrai of Harvard Medical School, who studies biomedical informatics. The AI, he says, has been especially helpful in flagging methodological problems and improving papers’ clarity. But “we are still verifying the claims they’re making and reading them against the paper—trust but verify is still our approach.”

“AI will almost certainly have an expanded role [in NEJM AI ’s reviews] in the future,” Manrai says. “In the not too distant past, it would have been quite unimaginable that one of the leading medical AI journals would be using AI in peer review at all. I think the norms are shifting.”

Researchers have high hopes that genetically engineered immune cells known as chimeric antigen receptor (CAR)-T cells can treat and even cure autoimmune diseases such as lupus and multiple sclerosis by slaying the errant immune cells driving the conditions. But the current methods for producing CAR-T cells, in which the cells are generated in a lab and then infused into patients, are slow and expensive, and the treatment itself can be dangerous. A new clinical trial may point to a better way.

Reporting today in The New England Journal of Medicine , researchers say they have induced patients to generate their own CAR-T cells . The treatment, which involves dosing people with a virus that genetically alters some of their T cells, didn’t cause severe side effects in the 16 patients tested and showed early signs that it restored their immune systems to normal.

“This is a step forward” for this approach, says rheumatologist Georg Schett of University Hospital Erlangen, who wasn’t connected to the research. “It’s a clear go signal for a further study.”

The standard procedure for making CAR-T cells starts when patients provide copious amounts of their own T cells. The cells are shipped to a specialized facility, where they are infected with a harmless virus whose genome carries DNA coding for a cell surface protein known as a CAR. The virus inserts this genetic material into the genome of the T cells, causing them to produce the CAR.

When the modified T cells are returned to the patient's bloodstream, the CAR directs them to attack body cells that carry specific proteins on their surface. Five of the approved CAR-T varieties, for example, target CD19, a protein on other immune cells known as B cells.

The approach has already notched successes in treating blood cancers. For instance, the cells have received approval for treating B cell acute lymphoblastic leukemia and large B cell lymphoma, two cancers in which B cell numbers explode. Certain B cells also drive autoimmune conditions by producing antibodies that attack the body’s own tissues. In hopes of replicating the experience with cancer, researchers have launched several dozen clinical trials to test the cells for autoimmune diseases such as myasthenia gravis and lupus.

But treatment with standard CAR-T cells is hard on patients—and on their wallets. The manufacturing process is complicated, time-consuming, and costly, averaging about $500,000. Moreover, patients often need chemotherapy to help the engineered cells establish themselves in the body.

There are other concerns as well. Pharma giant Novartis announced this week it had halted eight of its clinical trials of traditional CAR-T cells for autoimmune diseases because three patients died from a rare immune side effect.

Spurring patients to generate their own CAR-T cells inside their bodies, the so-called in vivo strategy, would spare them from chemo and should be faster and cheaper as well. To test this approach, a team led by neurologist Dai-Shi Tian of Tongji Hospital of the Huazhong University of Science and Technology used a virus that infects T cells in the body and endows them with a gene for a CD19-recognizing CAR. They infused the virus into patients with autoimmune diseases that affect the nervous system, including multiple sclerosis and myasthenia gravis.

The side effects of the treatment were limited. None of the patients developed a potentially fatal complication of CAR-T cell therapy that involves nervous system inflammation, for example. Although 11 of the recipients did develop a condition in which levels of immune molecules in the blood soar, they didn’t become severely ill, and their symptoms disappeared within 1 month.

The viruses used to engineer the T cells insert their DNA into the patients’ genomes, and some researchers worry this behavior could cause mutations or even awaken cancer-promoting genes. But when the scientists tested where the viral DNA ended up, they found it usually settled in sections of the genome that lie between genes.

The goal of CAR-T treatments for autoimmune disease is not to eliminate patients’ B cells for life. Instead, researchers think that by temporarily banishing the cells, they can reset the immune system so it no longer attacks patients’ own tissues. Tian and colleagues say they saw hints of that reboot. The number of B cells plunged shortly after the treatment but then rebounded in some of the patients.

Gerhard Krönke, an immunologist at the Charité University Hospital in Berlin who wasn’t connected to the trial, says it provides “a proof of principle that in vivo CAR-T cells can achieve effects similar to [labmade] CAR-T cells.”

After tracking the patients for a median of 6 months after the treatment, Tian and colleagues also uncovered preliminary evidence that the approach was easing symptoms. Patients with multiple sclerosis showed improved scores on a disability assessment, for example, and they didn’t develop any new lesions in their nervous systems. The study’s results suggest this approach for generating CAR-T cells “is clinically feasible and biologically active,” Tian says.

A clinical trial published last year also reported positive results for a different in vivo approach that doesn’t rely on viruses in patients with lupus. But the new study included more patients and tracked them for a longer period of time.

Still, the long-term benefits and risks of the treatment remain uncertain, notes Ken Young, a hematologist at Duke University Medical Center who studies CAR-T cells. “We still don’t know if the effect is sustainable.”

The deaths in the Novartis trials have also given the CAR-T field pause. No patients in the new study developed the side effect that spurred the company to stop its clinical trials. But this complication occasionally occurs in patients treated with conventional CAR-T cells, and in vivo approaches could also trigger it, Schett says. It’s likely to be very rare, however, he says.

The acceptable risks from CAR-T cells are typically much lower for autoimmune diseases than for cancer because patients are usually not as sick, says cell and molecular biologist Haig Aghajanian of the University of Pennsylvania Perelman School of Medicine. He worries that because the viral DNA remains in the patients’ genomes, B cells could continue to die long after the treatment. That effect could leave patients with low levels of the cells that are important for fighting infections, he says. “I wouldn’t have the guts to do it.”

LISTEN TO THIS ARTICLE

As so often happens with the Ig Nobel Prize, Ilona Croy initially believed the notice of her award was a scam. A neuroscientist at Friedrich Schiller University Jena, she was told she had won the olfaction prize for her study on parents who think their teenagers smell much worse than their babies . Still, she decided to accept, receiving a (worthless) $10 trillion Zimbabwean bill and a hand-built, fungi-themed trophy. It was all standard stuff for science’s silliest award ceremony—except this time it didn’t take place in the United States.

For the first time in the Ig Nobel’s 36-year history, winners accepted their prizes in Zürich, not in the Boston area. Croy is one of many awardees who support the change. Making the trip from Germany to Switzerland was an “easy” decision, she says. If the ceremony had been hosted in the U.S. as usual, “I would have thought twice,” she says, because of politics as well as distance. “The perception I have is that there is not so much of a welcoming culture anymore.”

Last year’s event was marred by the absence of nearly half of the 10 winning teams , for reasons including visa restrictions and the research and border policies of U.S. President Donald Trump. As a result, the Igs—whose motto is to “first make people laugh, and then make them think”—made the difficult decision in March to change its location indefinitely . “In order to do the ceremony and still have it feel like it always has at its core, we had to move it to a friendlier place,” says Marc Abrahams, founder and master of ceremonies of the Igs and editor of the scientific humor magazine Annals of Improbable Research .

Luckily, Zürich was eager to take on hosting responsibilities. Abrahams says cities including Paris, Berlin, Amsterdam, and Athens, Greece, also jockeyed for the honor. But the University of Zürich (UZH) and institutions of the ETH Domain, which includes ETH Zürich, offered institutional funding—something past U.S. hosts such as Harvard University and the Massachusetts Institute of Technology never did. Next year’s ceremony will be held in Antwerp, Belgium, the organizers announced at this year’s event. Every other year it will return to Zürich, rotating between other European countries in off years.

University of Oxford evolutionary biologist Matilda Brindle, who notched the biomechanics prize for devising a precise definition for kissing across animals including birds, ants, and polar bears , says the move abroad is a nice way to “reflect that science is an international endeavor.” It’s more central to the winning teams from Europe and Asia, and there are fewer visa restrictions. “I think it’s probably easier for people to come to Europe than the U.S. right now.”

That statement rings true for Tadd Truscott, a fluid dynamicist at the King Abdullah University of Science and Technology in Saudi Arabia, who won the physics prize for designing a splash-free urinal . Of his international lab group of 20 students, including people from Uzbekistan, India, and Nigeria, only two are now eligible to travel to the U.S. because of travel bans and visa difficulties, he says. One of his students was even held at the border for 10 hours without food or water, despite having a visa. “Unfortunately, it makes the U.S. feel like a noninternational place,” says Truscott, who lives in the U.S. for part of the year. Five of the other nine winning teams also include researchers based in the U.S.

One of Truscott’s co-winners, Kaveeshan Thurairajah, a fluid mechanics Ph.D. student at the University of Waterloo, notes that universities’ freedom and funding are in jeopardy in the U.S. “An attack on sovereignty is not something that a lot of us are comfortable with.”

For biology winner João Miguel Alves Nunes, a zoology Ph.D. student at the University of São Paulo, attending the ceremony is a professional statement—and a personal milestone. “As a Brazilian scientist, I feel proud to join the first ceremony outside the United States,” says Nunes, who won for stepping on venomous snakes to determine when they will bite . (The study was later retracted because of concerns about the stepping method.) “When visas or political problems stop scientists from meeting, science loses important voices,” he says. “The Ig Nobel’s move reminds us that science belongs to the world, not to one country.” The award also provided him the opportunity for his first-ever international trip and first airplane ride.

Although the move abroad is bittersweet, it has also reinvigorated the ceremony, Abrahams says. Tickets for the 1500-seat venue sold out within a few days, volunteers have been abundant, and the aging planning team has been finding new successors. The team has also had fun developing new, Switzerland-specific twists, including a yodeler, and university presidents donning banana costumes to usher off lengthy speakers. “It feels almost like it did at the very beginning.”

LISTEN TO THIS ARTICLE

A speck in a dish—that’s all the nervous system of the fruit fly Drosophila melanogaster looks like to the naked eye. But look closely enough and it resolves into a network of 166,700 nerve cells (shown above) connected through 124.2 million synapses. The intricate map —called a connectome—was published today in a package of four papers in Cell , marking the end of an ambitious 20-year research project.

Researchers published the connectome of the female fruit fly brain in 2024. This new and final instalment reveals the structure of a male’s brain—allowing comparisons between the two sexes—and goes a step further by including the ventral nerve cord. To generate the map, the team sliced the nervous system into manageable “slabs,” imaged the whole thing using a technology called enhanced focused ion beam scanning electron microscopy, then stacked images to create a 3D model. Machine learning helped trace and fill in the neurons.

The new papers use the map to begin answering questions about how the brain functions. For example, researchers compared the male and female brains and found that sex differences were concentrated in higher brain centers , whereas sensory and movement circuits were shared. In another paper, they traced the pathway of the entire taste system all the way from the sensory neurons to the motor neurons driving behaviors such as feeding and courtship.

LISTEN TO THIS ARTICLE

Once the media’s go-to scientist for all infectious disease stories, Anthony Fauci has largely avoided the spotlight over the past year—attracting the most press attention for repeatedly invoking the Fifth Amendment at a Senate hearing in July. But today he sat down for a nearly 2-hour interview with Lawrence Gostin, an expert in global health policy at Georgetown University, which hired Fauci in 2023 shortly after he left his longtime role as head of the U.S. National Institute of Allergy and Infectious Diseases (NIAID). Although the conversation, which was webcast, largely avoided politics, Fauci did address some hot-button issues surrounding the COVID-19 pandemic, vaccination, and global health.

Fauci never mentioned President Donald Trump or Senator Rand Paul (R–KY), two of his loudest critics, but he defended his actions during the pandemic. “I’m not sure we would change much” about the public health response, he said, contending there’s “corporate selective amnesia” about how dire the situation was in the first part of 2020. “What you see going on right now is all blame game,” he said. “That is not a constructive way to figure out how you can do better next time.”

Fauci has taken a good deal of heat for schools and businesses closing down, recommendations about masking and vaccination, social distancing measures, and the origin of the pandemic itself. “I’m the bad guy,” he joked. “I created the virus. I shut the schools. You know, I did everything, I did everything. Throw away the key.” (Paul indeed has argued that Fauci should be prosecuted for his pandemic actions, but shortly before leaving office, then-President Joe Biden issued him a preemptive pardon.)

Fauci stressed that he did not set policy; that was up to governors, mayors, and teachers’ unions. He told Gostin he wanted schools to reopen more quickly than they did, and he contested the claim that social distancing measures—such as the “30 days to slow the spread” invoked by the first Trump administration—were lockdowns. “We didn’t lock anything down,” he said. “Were you locked in your house? I wasn’t locked in my house. In China, you were.”

Fauci’s critics have also charged him with covering up what they believe was the origin of the pandemic: a leak of SARS-CoV-2, the virus that causes COVID-19, from a laboratory in Wuhan, China, that received some funding from NIAID. Fauci said he has not ruled out a lab leak. “I keep a completely open mind,” he said, “but I’m a scientist, and I go by data.” A half-dozen papers in high-profile journals by “highly qualified evolutionary virologists” support the alternative theory, that the virus naturally jumped into humans from animals. “The people who say it definitively came from a lab leak, I’m sorry, but there’s nothing in the data that backs that statement,” he said.

The changing advice from Fauci and other leaders during the early months of the pandemic has also been fodder for critics. That was unavoidable when the science was changing fast, he argued. “One of the things that we’ve got to articulate better is to get people to understand that science is a process that gathers information that is available to you at a given time,” he said. “Maybe the arithmetic science is absolute, so that in January of 2020, two plus two equaled four, and in September of 2026, two plus two still equals four. But our understanding of the virus in January of 2020 is remarkably different than our understanding now.”

He also commented on more recent issues. The current Trump administration’s decisions to cut funding for global health and drop out of the World Health Organization (WHO) were both mistakes, he said. “Global health is absolutely critically important because of a simple fact that by the very nature of the 21st century, we live in a global community,” he said, calling WHO “imperfect” but “extremely valuable.”

He singled out the rapid spread of Ebola in the Democratic Republic of the Congo (DRC) now taking place. “Ebola in the current outbreak in the DRC was very likely going on for several weeks before anybody even noticed it, and the reason is we didn’t have the on-the-ground capability,” he said, making an oblique reference to the Trump administration eliminating the U.S. Agency for International Development and paring back overseas fieldwork done by the Centers for Disease Control and Prevention.

He strongly defended COVID-19 vaccines, which the current secretary of Health and Human Services, Robert F. Kennedy Jr., has called “the deadliest vaccine ever made.” Kennedy has specifically assailed the messenger RNA (mRNA) technology that is the basis of the COVID-19 shots most widely used in the United States, and Fauci lamented that the message has gained traction. “This acceptance that mRNA vaccines kill more people than COVID does, the number of people that believe that is really quite frightening,” Fauci said.

The data showing the mRNA COVID-19 vaccines are safe and effective are “overwhelming, yet the new thing now is, ‘Oh, that’s fake news,’” he said. “If somebody really, really is convinced that it’s fake, nothing you’re going to say is going to convince them otherwise.”

Similarly, he assailed a theory pushed by Kennedy that children receive too many vaccines at the same time. “The idea that if you give a lot of vaccines, overwhelming the immune system, is just scientifically not the case,” he said. “What that doesn’t take into account is that when your child goes to a day care, it is a petri dish of infections.”

Gostin, who took questions from the small audience and some people online, closed the discussion by praising Fauci. “You’re a national treasure,” he said. But Fauci has faced death threats, and the Georgetown discussion, originally billed as open to the public, was closed for security reasons.

LISTEN TO THIS ARTICLE

In ancient Rome, shepherds feared the appearance of bumpy pustules on their sheep—a telltale sign of deadly sheeppox. Even today, the disease poses a major threat, capable of wiping out entire flocks. Yet the disease’s origins—and historical impact—have remained mysterious.

In a study published today in Science Advances , researchers pin the virus’ emergence to about 5000 years ago , roughly the same time people started to intensively shear sheep for wool. “I was so jealous when I saw this paper,” says Floris Breman, an evolutionary biologist at the Belgian national health agency Sciensano, because he wished he could have made the discovery. “It’s very important because it shows us how viruses arrive in a population.”

The peek into the pathogen’s past came from an unusual source: dozens of medieval books and legal documents written on parchment, which is made from specially treated animal skins. “It turns out parchment is an extremely good preserver of this type of biological data,” says University of Leicester historian Joanna Story, a co-author of the new study.

As part of a larger look at the origins of medieval manuscripts, researchers rubbed pages of the books with erasers, then extracted DNA from the eraser rubbings. That there was DNA to recover at all came as a surprise: Before becoming pages in a book, cow, goat, and sheep skins were soaked for days in pits filled with lime, a caustic chemical, and then painstakingly scraped and stretched by hand. “It’s a pretty disgusting process,” says University College Dublin geneticist Kevin Daly, who led the new study.

Finished manuscripts were often written on and handled for centuries, giving DNA trapped in the parchment ample time to degrade or become contaminated.

Yet even after centuries on the shelves of monasteries and archives, the manuscripts contained enough intact DNA to identify the species skinned to make the parchments as well as viruses and other pathogens that afflicted the animals, including dozens of extinct sheeppox strains. “I couldn’t have imagined they’d get all this data from parchment,” says Nicolás Rascovan, a geneticist at the Pasteur Institute who was not part of the research team. “They’ve discovered a new source of ancient pathogen DNA that will be very helpful to other teams in the future.”

That’s because a single book amounts to a population sample. The average medieval manuscript was a huge investment, requiring anywhere from a few dozen animals to hundreds of skins for the largest Bibles. “When you pick up a book, you’re picking up a flock of sheep or herd of cattle,” Story says.

Though not as well-known as its distant cousin, smallpox, farmers in Europe and Africa considered sheeppox a major threat. A sheeppox outbreak can kill between 50% and 80% of a flock’s young. The virus is highly contagious, and if not caught quickly can devastate herds across entire regions and cause hundreds of millions of dollars in damage. “It’s a big deal,” Breman says. “If you have 50% to 80% of your lambs dying, you have no new sheep for the year. The impact is huge.”

But how long has this disease been a shepherd’s bane? Comparing differences in various strains taken from medieval manuscripts dating from between 700 C.E. and 1450 C.E., the team estimated the samples all had a common ancestor that evolved into a sheep-specific disease about 5000 years ago.

That’s roughly when archaeological evidence, such as loom weights, points to people on the Eurasian steppes beginning to breed and manage sheep for their wool. Bigger flocks kept in proximity would have been fertile ground for a virus to specialize in these animals. Highly mobile Bronze Age shepherds would have spread widely and rapidly. “If sheep are being more intensively managed and bred to produce wool,” Daly says, “that might have acted as a catalyst for the virus to spread or evolve.”

Royal Irish Academy collections manager and archivist Alison Ray suggests medieval monks probably knew what they were doing when they turned pox-riddled sheepskin into pages. “If a whole herd goes, you can’t eat them,” she says, “but using diseased skins for books is a practical choice.”

LISTEN TO THIS ARTICLE

Next month, NASA’s New Horizons mission plans to shut off two instruments that study the solar wind because of a lack of funding from the agency, mission scientists say. The move comes even as the spacecraft nears the outer edge of the Solar System, following a different route than its far-out predecessors , Voyager 1 and 2.

Best known for its flyby of Pluto and exploration of the Kuiper belt , a region of small icy bodies, New Horizons has also provided a continuous look at the “wind” of charged particles the Sun emits and the vast bubble it creates around the Solar System. Much of this work is driven by two particle detectors: the Pluto Energetic Particle Spectrometer Science Investigation, which measures energetic particles, and Solar Wind Around Pluto, which measures the lower energy particles of the solar wind. But NASA has withheld $2 million in funding to support the instruments for each of the past 2 years, money that covers the salaries of engineers and scientists who download and analyze the data. The mission has been left with no choice but to shut them off, says Alan Stern, New Horizons’s principal investigator and a planetary scientist at the Southwest Research Institute. “We just can’t afford to keep the lights on on the heliospheric instruments.”

Shutting off these instruments “would be an incalculable loss,” says Gary Zank, a space physicist at the University of Alabama in Huntsville. Unlike the Voyager missions, New Horizons can directly measure “pickup ions,” neutral atoms from interstellar space that become charged inside the heliosphere and get swept up by the solar wind. These particles sap energy from the solar wind and play a role in what happens at the termination shock, where the supersonic solar wind abruptly slows.

New Horizons is currently 66 times farther from the Sun than Earth and twice as far away as Pluto. It could cross the termination shock in the next few years. The spacecraft is in perfect health and could operate until 2050, Stern says. Such longevity would allow it to eventually cross the heliopause into interstellar space, enabling insights into how that boundary—which marks the edge of the Sun’s bubble—can shrink and grow over time, says James Green, NASA’s former chief scientist, who organized a petition to keep the instruments running. “This is a unique opportunity that we’re not going to get very often.”

Earlier this decade, New Horizons’s future was in doubt, as different divisions of NASA’s science directorate tussled over who should fund it. The agency eventually struck a deal in which planetary science would pay the bulk of operations, some $10 million annually, while heliophysics would provide $2 million each year. The heliophysics division provided that money for a couple years, but it has not done so for the 2025 or ’26 financial years, necessitating the shutoff. The agency has provided no explanation for the withholding, says Fran Bagenal, a planetary scientist at the University of Colorado Boulder and mission team member. “Just arbitrarily turning things off here and there is short-sighted.”

A reversal of the decision may be coming, however. In a social media post today , NASA Administrator Jared Isaacman said the agency was looking into the problem. “The big dollars were already invested in New Horizons, so generally supportive of generating high-quality scientific returns for low-dollar spend,” he wrote. (In response to questions, a NASA spokesperson referred to the administrator’s post.)

Such a reversal would be welcome news, given the essential role that New Horizons is playing in studying the Sun’s distant influence—and the physics that shapes the environment around other stars, Zank says. “It beggars belief that people are even contemplating shutting this down.”

LISTEN TO THIS ARTICLE

Several clinical trials have recently sprung up to test an appealing prospect: that delivering electrical or magnetic energy to the brain from outside the skull could ease burdensome symptoms in women, such as postpartum depression or pain related to menstrual cycles or endometriosis. Noninvasive brain stimulation treatments have been approved for a handful of mental health conditions and explored in many other disorders. But until recently, women’s health was not a focus of this research.

This is starting to change, as work highlighted at a session of a neuromodulation conference last month in New York City made clear. “We are thinking more about issues specific to women and how neurotech can meet those needs,” says psychiatry researcher Giuseppina Pilloni of New York University Langone Health, one of the conference’s organizers, who is not involved in the new trials.

Researchers are testing whether transcranial magnetic stimulation (TMS) can relieve depression symptoms during and after pregnancy, and whether it might even be used to prompt release of reproductive hormones. Other groups are seizing on a less proven but more affordable and accessible technology that can be used at home: transcranial direct current stimulation (tDCS). Both approaches are yielding encouraging hints of benefit.

TMS, which delivers strong magnetic pulses that can boost or inhibit neural firing, is approved in the United States for depression, migraine pain, obsessive-compulsive disorder, and smoking cessation. But safety concerns about running trials in people who are pregnant or lactating, along with the logistical challenges of accommodating them, have left the technology understudied in these groups, says psychiatrist Veerle Bergink of the Icahn School of Medicine at Mount Sinai. “It’s easier to do this in almost every other population,” she says. “But the fact that it’s difficult to do these studies doesn’t mean they shouldn’t be done.”

Bergink and her colleagues are investigating an intensive 5-day regimen of TMS targeted to the dorsolateral prefrontal cortex, a brain region thought to have a role in depression, in about 200 women who have recently given birth . (A small, nonrandomized trial by a different group is enrolling both pregnant and postpartum participants to test the same accelerated TMS regimen.)

The value is clear: Antidepressant medication takes time to work, “and there’s time pressure when there’s a baby,” Bergink says. “This [intervention] is something we can deliver in 1 week.” People may also wish to avoid exposing their newborns to the trace amounts of antidepressants in breast milk, she notes, even though the most widely prescribed drugs, selective serotonin reuptake inhibitors, are generally considered safe during nursing and even pregnancy.

Catarina Ferraz, a biomedical engineer at the City College of New York, has been developing a more exploratory use of TMS: modulating the activity of hormone-producing neurons in the hypothalamus, a deep-brain structure, to prompt release of lutenizing hormone (LH) by the pituitary gland at the base of the brain. This could potentially lead, she thinks, to nondrug treatments to prompt ovulation, or for infertility or endometriosis, two conditions linked to disruptions in LH timing and levels.

At the conference, she showed results from a pilot study of TMS designed to influence hypothalamic neurons. The treatment increased blood levels of LH beyond expected changes in five of seven healthy women receiving the stimulation. Ferraz is now planning to try the intervention in a group of people with amenorrhea, or a missing period.

Emily Jacobs, a neuroendocrinology researcher at the University of California, Santa Barbara, cautions that only one patient in Ferraz’s study had what she considers a robust response to the stimulation. Still, the concept of using a piece of technology to drive changes in the endocrine system is “so cool in theory,” she says. “I’ll await more data to decide if they’ve stumbled on something truly remarkable.”

TMS treatments involve bulky equipment and require experts to administer, making them suited only for use in clinics. In contrast, tDCS uses portable headgear to deliver low-voltage current to parts of the cortex, or outer layer of the brain. Unlike TMS, the current is not strong enough to initiate or suppress neural firing, but it can influence the firing patterns of already active neurons, making them more or less excitable depending on how electrodes are placed.

The only indication for tDCS approved by the U.S. Food and Drug Administration (FDA) is to treat depression . Its modest effects compared with placebo have prompted a group of scientists to argue in a recent editorial that the approval late last year may have been premature. But researchers at the conference showed preliminary evidence that its benefits could be improved with more frequent use, and others are hopeful its indications will soon expand. With the FDA approval, “we’ve broken the sound barrier” for clinical applications of tDCS, neurologist Mark George of the Medical University of South Carolina told attendees.

Pregnant women with depression were not covered by the approval because they were excluded , along with breastfeeding women, from the key clinical trial. But psychiatrist Simone Vigod of the Women’s College Hospital in Toronto thinks tDCS may be a good option for them, including, in some cases, as an alternative to antidepressants. “I’ve seen so many pregnant people struggle” with concerns about taking medication , Vigod says. “It would be wonderful to have another effective option to add to our toolkit.” Twenty years of safety data paint a reassuring picture for tDCS, she says: “It only hits the area of the brain it’s aimed at,” and does not impact blood pressure or heart rate, which must be kept within safe ranges during pregnancy.

Vigod is leading a large randomized, controlled trial to determine whether a course of treatment with remotely supervised at-home tDCS can not only reduce pregnant participants’ depression symptoms, but also lower their risk of developing postpartum depression after they give birth.

A headbandlike tDCS device was cleared last year in Europe to treat a different concern: pain and mood symptoms associated with menstruation. Marketed to women by Samphire Neuroscience, it is also sold over the counter in the U.S. as a “consumer wellness device.” It delivers current to the dorsolateral prefrontal cortex, like the approved tDCS device, and one additional area: the motor cortex, a brain region involved in pain.

The headband was cleared in Europe based on results from a trial designed to measure its ease of use and evidence from prior studies on tDCS. This year, Brazilian researchers working with the company reported inconclusive benefit from a trial in 40 women with dysmenorrhea, or period pain. Samphire is now sponsoring much larger randomized trials for dysmenorrhea and for endometriosis pain. At the New York conference, Samphire head Emilė Radytė presented findings suggesting the headband helped in premenstrual dysphoric disorder, a condition marked by severe mood shifts.

Jacobs says there’s still a lot of work to be done to ensure such interventions are truly safe and effective—“you don’t want to go around arbitrarily zapping your brain,” she says—but the momentum in the field is welcome. “It’s definitely encouraging to see the neuro stim space beginning to ask questions that might actually be tailored to women’s bodies.”

LISTEN TO THIS ARTICLE

The Australian Antarctic Division (AAD) has made an unprecedented and controversial decision to evacuate its research station on a remote sub-Antarctic island for fear that a highly pathogenic strain of H5N1 avian flu known as 2.3.4.4b will soon reach its shores. The risk to human health on Macquarie Island—located in the Southern Ocean midway between Australia and Antarctica—is low, the agency said when announcing the move last week. But the virus could wipe out thousands of southern elephant seal pups born on beaches next to the station, potentially leading to psychological distress for staff.

Officials announced they will send Australia’s sole polar vessel to remove all 24 staff from the island in early October, ahead of the summer field season. “It’s amazing that they are thinking about the researchers’ well-being,” says Roxanne Beltran, a biologist at the University of California (UC), Santa Cruz who studied an H5N1 outbreak in northern elephant seals in California earlier this year.

But because in-person research will be shut down indefinitely, until the station reopens, some scientists say valuable data won’t be collected. “This decision raises a much bigger question about how we conduct science in an era when major wildlife disease outbreaks may become part of the Southern Ocean environment,” says Jane Younger, an Antarctic disease ecologist at the University of Tasmania. “If the response to that risk is to withdraw scientists whenever a serious outbreak becomes possible, that has profound implications for our ability to understand and respond to future ecosystem change.”

It could also be a lost opportunity to study the initial stages of an H5N1 outbreak in detail right at its epicenter, notes Claudio Campagna, a researcher who studied southern elephant seals in Argentina for decades with the Wildlife Conservation Society. “This is a chance to make a huge contribution to epidemiology,” he says.

The highly pathogenic strain has been spreading around the world since 2020, infecting wild and domestic birds. Although the virus can infect many kinds of mammals, rarely does it spread within species. But in November 2022, thousands of sea lions began dying in Peru. The next year even more southern elephant seals perished in Argentina. The mortality rate reached 95% among pups, which were struck down with tremors and convulsions. Campagna says he was barely able to watch.

By October 2023 the virus had arrived in the sub-Antarctic region, home to huge populations of southern elephant seals. Aerial surveys conducted by the British Antarctic Survey suggested about 53,000 died on South Georgia and the South Sandwich Islands, nearly halving the breeding population there. In October 2025, H5N1 was detected hitting southern elephant seals on Heard Island and the McDonald Islands—an external territory of Australia—leading to the death of three-quarters of the pups, according to a preprint posted on bioRxiv. Seabirds, some of which may scavenge on the drifting carcasses of infected marine mammals, are thought to carry the virus long distances.

Macquarie lies thousands of kilometers east of Heard and the McDonald Islands, but the government said expert assessments suggest a “very high likelihood” that H5N1 could reach the island during the 6-week elephant seal breeding season, which starts next month. Amandine Gamble, a disease ecologist at Cornell University who was not involved in the assessment, thinks it is “more likely than not” that the virus will arrive this year. But, she adds, “Making very accurate predictions for one site and one species is still something that we’re not very good at.”

The area around the research buildings is home to a colony of about 19,000 breeding females that has been studied since the station was founded in 1948. Each year, research staff weigh pups and fit them with satellite tags—providing important information about population trends and how they shift with changing environmental conditions. Researchers also monitor breeding populations of seabirds, such as the grey petrel, and have studied their recovery after a 7-year effort to remove invasive rabbits and rodents from the island was completed in 2014.

A government spokesperson told Science AAD would not reveal how many researchers are currently on the island, and it did not say whether any were given an option to stay. That irks Younger, who says alternatives might have included a smaller specialist team, shorter deployments, and additional psychological support. “I don’t think the government has yet explained how it concluded that removing everyone from Macquarie Island was the best, or only, option.” Other nations’ programs in the sub-Antarctic monitor bird flu outbreaks under enhanced biosecurity, she points out.

When announcing the decision, the agency also cited the risk of logistical problems if an unplanned evacuation were to be needed later on. Diverting Australia’s only polar vessel on short notice to remove staff from Macquarie would disrupt operations at the country’s three Antarctic stations, the agency said.

“It’s a decision that makes a lot of sense,” says Marcela Uhart, a wildlife veterinarian at UC Davis. She agrees that an emergency evacuation would be disruptive and adds that virus shed by infected animals could threaten the station’s water supply.

Uhart witnessed catastrophic losses of elephant seals in Argentina, so she also understands the psychological toll they can take on researchers. “Nothing prepares you for thousands and thousands of animals that are dying in a very short time.” Younger agrees that mass mortality can be deeply distressing, but adds: “For many scientists, being told to turn away and ignore what is happening in the ecosystem we’ve dedicated our careers to is much worse.”

LISTEN TO THIS ARTICLE

On 23 November 2023, a ripple in spacetime swept across Earth. Lasting just one-tenth of a second, the signal was picked up by the two U.S.-based detectors that make up the Laser Interferometer Gravitational-Wave Observatory (LIGO). The source of the wave appeared to be a violent merger of two rapidly spinning black holes, each more than 100 times the mass of our Sun. But LIGO researchers cannot explain how such hefty black holes can form, nor why they would be spinning so fast.

A paper published last week in The Astrophysical Journal Letters proposes an intriguing possibility: It could be an illusion . A less extreme black hole merger, magnified by an intervening object acting as a gravitational lens, could produce the signal LIGO saw, known as GW231123. “It’s a very interesting possibility,” says cosmologist Valerio De Luca of the University of Pennsylvania, who was not involved in the work. “The question is to understand which kind of lens could work.”

In partnership with the Virgo detector in Italy and KAGRA in Japan, LIGO has detected hundreds of gravitational wave events over the past decade, mostly the mergers of black holes with at most a few tens of solar masses. Black holes are thought to form when giant stars run out of fuel and collapse on themselves. But theorists think truly massive stars end their lives differently: in a pair-instability supernova, in which the core becomes so hot that the gamma rays it produces transform into particles. The sudden loss of radiation pressure causes the core to collapse, triggering an explosion that leaves nothing behind. At still higher masses, the star can again collapse directly into a black hole. The result is a predicted gap, in which black holes with masses between 60 and 130 times the Sun’s mass should not exist.

That’s why GW231123 is a puzzle. The shape of the wave detected by LIGO fits a merger of black holes with 137 and 101 solar masses combining to form a 230–solar mass black hole. And although black holes are expected to spin, the two in GW231123 would be extreme, rotating at about 90% and 80% of the theoretical maximum.

In dozens of papers and preprints, theorists have conjured up alternative explanations, including primordial black holes created soon after the Big Bang, fault lines in the fabric of space known as cosmic strings, and even a wormhole, whose gravitational waves would repeatedly echo from inside it. “This is the event that just keeps giving,” says astrophysicist Nelson Christensen of the Côte d’Azur Observatory (OCA).

The most straightforward explanation is that the two black holes were themselves the products of earlier mergers, but there are problems with that scenario. Multiple mergers would likely be required to reach such extreme masses. And when such mergers occur, the resulting black hole is often given a shove that would send it flying out of dense stellar clusters, making it less likely to encounter other black holes and merge again.

Srashti Goyal of the Max Planck Institute for Gravitational Physics wondered whether a gravitational lens might instead be responsible for GW231123’s apparent extremity. Astronomers have known for more than a century that the gravity of a large foreground mass can bend the path of light from a background source, magnifying its appearance. Such lenses can magnify gravitational waves, too, Goyal says.

But lensing of gravitational waves is more complicated because their wavelengths are so long, tens of thousands of kilometers in the case of a heavyweight black hole merger. Whereas a galaxy is large enough to bend them like an optical lens, smaller objects such as stars or black holes would diffract them instead, scattering the waves to produce a messier, more complicated signal.

Goyal and her colleagues built a model to account for both effects: a compact lens, such as a midsize black hole, embedded in a larger, extended lens, such as a galaxy. They calculated what combination of black hole merger and lens masses and positions could produce a signal like GW231123 and found that, if magnified, a far less massive black hole merger would work. “When we do the matching, we can statistically say which hypothesis is a better fit,” Goyal says. “It turns out lensing is a better fit.”

Their model suggests the signal could come from a merger of black holes with a combined mass of 140 Suns, and without the extreme spin rates. The merger would be much more distant—beyond the range of the LIGO detectors—but its signal would be magnified by a roughly 500–solar mass black hole surrounded by an extended lens. Though not a definitive smoking gun, it may be the most persuasive explanation of GW231123 so far, says OCA’s Iuliu Cuceu. “This probably is the best explanation of the data we have.” But he has reservations about drawing too many conclusions about GW231123. “This event is so exceptional that we do not know what is normal.”

Given that LIGO only picked up a brief part of GW231123’s signal—five wave cycles—researchers may never be able to say with certainty what its source was. But LIGO and its counterparts in Italy and Japan, currently shut down for upgrades, will soon start to observe again with greater sensitivity. More extreme events, including lensed ones, will put these models to the test. “It’s hard to say what the odds are, but I think it’s exciting that we are already seeing some things which are not really explainable … without lensing,” Goyal says.

LISTEN TO THIS ARTICLE

Don’t get too excited, but the world’s largest dark matter detector has spotted a single unusual particle. Announced today at the TeV Particle Astrophysics conference in Japan, the event recorded by the LUX-ZEPLIN (LZ) detector could mark the first detection of a particle of dark matter, the mysterious invisible stuff whose gravity appears to bind the galaxies. However, it’s far too early to claim a discovery, physicists warn. And the dark matter particle, if real, wouldn’t be what many theorists expected.

“How do you even make sense of one event?” muses Tom Shutt, a particle astrophysicist at SLAC National Accelerator Laboratory and co-founder of the LZ project. “We just decided we should publish and think really, really, really hard about what that event could be.” The hint of something revolutionary will tantalize physicists, says Wick Haxton, a theoretical physicist at the University of California (UC), Berkeley. “It’s like seeing a present under the Christmas tree. You’re hoping it’s something good and you’re excited no matter what.”

The ways in which galaxies whirl and gather suggest they reside in much larger clumps of dark matter. Dark matter can’t be seen directly because it doesn’t absorb or emit electromagnetic radiation, yet studies of the afterglow of the Big Bang—the cosmic microwave background radiation—indicate it makes up roughly 85% of the matter in the universe . But no known particle is both sufficiently hefty and inert to explain it.

In the 1980s, theorists realized they might explain dark matter by hypothesizing just one more type of particle. It would have to be stable and about 100 times as massive as a proton, and it would have to interact only through gravity and the incredibly feeble weak nuclear force. Such weakly interacting massive particles (WIMPs) would naturally linger from the Big Bang in just the right amount to account for dark matter, provided that, like the photon, such a particle was its own antiparticle. These particles also appeared naturally in theories built around a concept called supersymmetry. So they became the main quarry for dark matter hunters.

WIMPs should be floating all around us, and every now and then one should bump into an atomic nucleus and send it recoiling. Physicists once thought a detector weighing less than 1 kilogram could reveal such nuclear recoils. But a series of ever bigger detectors has failed to spot WIMPs. LZ, the current biggest WIMP detector , contains 7 tons of frigid liquid xenon, providing an astronomical number of targets for passing WIMPs.

The detector lurks 1480 meters deep in the Sanford Underground Research Facility, in a former gold mine in South Dakota. It started taking data in 2021 and still hasn’t seen what physicists expected. WIMPs should move quite slowly, only about 300 kilometers per second or 1/1000th the speed of light, so a xenon nucleus struck by a WIMP should recoil with low energy. Last year, LZ researchers looked for nuclear recoils between about 5 and 55 kiloelectron volts (keV) and saw nothing.

Now, however, the team has extended its search up to recoil energies of 270 keV and spotted a single event. Try as they might, LZ physicists could not account for it as a “background” event from, say, a stray neutron or gamma ray. So they decided they had to report it to the community, says Rick Gaitskell, a physicist at Brown University and spokesperson for the LZ collaboration. The LZ team will submit a paper on the work to Physical Review Letters .

But if the event was caused by a dark matter particle, the high recoil energy of 248 keV suggests it wasn’t an ordinary WIMP, Gaitskell says. Lower energy WIMPs should greatly outnumber higher energy ones, he explains. If the supposed particle interacts with nuclei in the simplest way, physicists should also have detected thousands of lower energy events, Gaitskell explains.

Since none have been seen, physicists will have to invoke more complex models in which the rate of interaction increases with the energy of the collision. “What you’re seeking to do is to suppress the low-energy guys, which would then explain why you see an event at the higher energy but not this much larger number of lower energy events,” Gaitskell explains.

Such behavior could require dark matter to be more complicated than just a single new type of particle . For example, the dark matter particle might have some internal structure, like an ordinary atom, so it would only interact if hit hard enough to excite it to a higher energy internal state. Or the “coupling” of the particle to ordinary matter could depend on its momentum, perhaps because it was mediated by another type of new particle.

Such models are not as contrived as they might sound, notes Jonathan Feng, a theorist at UC Irvine. “I’m really happy they did this search because there are whole classes of theories that have this momentum dependence,” he says. JiJi Fan, a theorist at Brown, says some of these models have been around for decades. “Even this one event could narrow down the parameter space” for such theories, she says. “That’s exciting.”

The most important question is whether the putative signal will persist as more data come in. LZ physicists estimate there’s about a one in 200 chance the event is a statistical fluke—far from the one in 3.5 million required to claim discovery. And the field of dark matter searches is strewn with dubious claims and anomalies that have faded.

If the new result is real, more signals should emerge soon. LZ researchers have already collected three times as much data as they used in the paper. And other detectors, such as XENONnT in Italy’s Gran Sasso National Laboratory, which is nearly as big as LZ, or the far bigger next version of the PandaX detector, currently under development in China, should be able to search for the high energy events, too.

“We can perform this study independently with a blind analysis to validate or invalidate LZ’s claim,” says Elena Aprile, a physicist at Columbia University and spokesperson for the XENON team. For dark matter hunters, Christmas is coming, although exactly when and how merry it will make them remains to be seen.

LISTEN TO THIS ARTICLE

Over the next 18 months, California voters will have two chances to approve spending billions of dollars to bolster the state’s scientific enterprise in the wake of cuts to federal research by President Donald Trump.

Voters in the ultrablue state have previously backed billions for stem cell science after a Republican president cut the field’s funding. But the fates of the two new initiatives—one going on the ballot in November, the second appearing in March 2028—are far from certain, especially as conflict-of-interest questions swirl about this year’s proposal, which would raise more than $8 billion for development of immunotherapies.

Yesterday, the California legislature passed a bill to invest $7.5 billion to create a state foundation for science and health research. The entity, if voters approve bonds dedicated to it, would make grants selected by peer review to California academic institutions across all fields, although biomedical research would likely be its top priority. Governor Gavin Newsom (D) is expected to sign the bill, which would make California the first state to broadly fund science at such a massive level.

But instead of putting the legislative initiative, SB895, to voters during the upcoming state elections on 10 November, lawmakers decided to delay the ballot proposal until California’s 2028 presidential primary. The bill’s chief sponsor, state Senator Scott Wiener (D), says supporters wanted to avoid a “crowded” November ballot.

That crowd includes a citizen-backed initiative, called Proposition 38 , that asks voters this fall to approve $8.4 billion in state bonds to fund immunology and immunotherapy research targeting diseases such as cancer and Alzheimer’s. The proposal was crafted by billionaire surgeon and medical inventor Gary Michelson, who also financed the campaign to collect the 546,000 signatures needed to get it on the ballot.

But Proposition 38 has sparked controversy because only half of the bond money would be allocated via a competitive, peer-reviewed process. Some $4.2 billion would go to a single unnamed nonprofit entity that, among other criteria, must be affiliated with a University of California (UC) medical center that serves at least 3.5 million patients annually and has received at least $250 million in philanthropic support. According to reporting by CalMatters , the only qualifying recipient would be the California Institute for Immunology and Immunotherapy (CIII), which is linked to the UC Los Angeles (UCLA) medical school. The institute has yet to open, but Michelson himself gave it $120 million in seed money in 2022. CIII has also received $200 million from the state to purchase a vacant shopping mall near the UCLA campus.

David Panush, a 35-year veteran of California state government, argues that when the state’s leading research universities and medical centers are reeling from funding cuts by the Trump administration, asking taxpayers to support only a tiny slice of that community doesn’t make sense. “The science community should be absolutely outraged,” says Panush, now a consultant on health care policy. “If you try to use the political process to benefit those doing just one type of research, you’re adding to the growing erosion of public trust in science.”

Health psychologist Robert Kaplan, a senior scholar at Stanford University’s Clinical Excellence Research Center and former head of the National Institutes of Health’s Office of Behavioral and Social Sciences Research, has joined Panush in mounting a shoestring effort to defeat Proposition 38. “There is a strong case for investing more in biomedical research, but [Proposition 38] is not the right way to do it,” Kaplan says. Both he and Panush prefer the route Wiener took in gaining the support of elected public officials before going to the voters.

The Yeson38 website notes the bill specifies that the single institute receiving half the bond raise “will be selected by the California Department of Public Health.” A spokesperson told CalMatters that “whichever institute is selected [will be] prepared to accelerate lifesaving research immediately while maintaining the highest standards of scientific quality, accountability, and stewardship of public funds.”

Proposition 38 has been endorsed by several patient advocacy groups, especially those focused on chronic diseases. However, the only national organization of researchers backing Proposition 38 is the American Association of Immunologists.

By comparison, the broader science foundation proposed in the bill approved this week is being championed by such research heavyweights as the UC system, whose 10 campuses would presumably be major beneficiaries, as well as the 60,000-member United Auto Workers union that represents many of their employees. (The UC system takes no position on Proposition 38 because it is legally not allowed to weigh in on ballot issues, unlike legislation, a spokesperson explained.)

Last week, Nobel Prize winners Jennifer Doudna and Randy Schekman, both at UC Berkeley, added their voices of support for the broader science and health foundation. “SB895 represents an important investment in the research infrastructure that will allow California to continue advancing scientific breakthroughs,” asserted Doudna, a biochemist who won the prize in 2022 for her co-discovery of the genome-slicing technique known as CRISPR.

Under SB895, the governor would appoint a majority of the foundation’s oversight council. The bill also includes language to spur commercialization of any breakthrough discoveries and requires “a portion of the revenues from such highly profitable inventions, including blockbuster drugs” to be returned to the foundation.

California has led the way with state-sponsored science. In 2004, voters approved $3 billion in state bonds to launch the California Institute for Regenerative Medicine, a response to efforts by then-President George W. Bush’s administration to block research on embryonic stem cells, and gave it another $5.5 billion in 2020. Texas has similarly raised $6 billion in bonds for cancer research and $3 billion for Alzheimer’s studies.

Wiener, who has not taken a position on Proposition 38, says he’s not worried the two science funding initiatives will cause voter fatigue. “That ballot measure [Prop 38] primarily funds one particular institute,” Wiener said last week at a briefing for reporters. “It’s just very different” from the foundation his bill would launch. “I don’t view it as a competitor.”

Every year, tens of thousands of pregnant women die from the blood pressure disorder preeclampsia, and many more have babies who are stillborn or die young after failing to grow properly in the womb. Now, researchers say they have a way to make these conditions easier to predict, potentially helping problem pregnancies get treatment sooner.

In a study published today in Nature Medicine , scientists who analyzed blood samples from thousands of pregnant women found that low levels of a protein called ISM2 during the first trimester were associated with higher risk of both preeclampsia and fetal growth restriction later on. Follow-up lab studies of placental cells showed ISM2 likely plays a causal role in the disorders, meaning the findings could help in the development of treatments, too.

“Finding new biomarkers that can not only predict preeclampsia, but also predict fetal growth restriction is phenomenal,” says Cathy Cluver, a clinician scientist at Stellenbosch University who was not involved in the work. “The fact that it was backed up by the preclinical work that they did was really exciting.”

Although both are diagnosed later in pregnancy, preeclampsia and fetal growth restriction are linked to problems with early development of the placenta, the temporary organ that sustains a fetus. In a healthy pregnancy, specialized placental cells called extravillous trophoblasts (EVTs) burrow into the uterus and help set up the blood flow between mother and fetus. If this process fails, the lack of oxygen and nutrients to the placenta can limit the fetus’ growth and trigger damage to the mother’s blood vessels, leading to high blood pressure.

Clinicians currently use a combination of approaches—often incorporating a patient’s family history and age as well as measures of certain blood proteins and ultrasound imaging—to try to identify those at high risk of these conditions. But the methods don’t predict every case, and some elements, such as ultrasounds, can be particularly difficult to access in low- and middle-income countries. Large-scale efforts supported by the World Health Organization, the Gates Foundation , and others are now underway to identify better biomarkers and therapeutics.

In the current study, funded by Wellcome Leap , a U.S. nonprofit launched by the U.K. charity Wellcome Trust, researchers at the University of Cambridge and collaborators analyzed data from a large project called the Pregnancy Outcome Prediction Study (POPS). The more than 4500 U.K. women in this study gave blood samples at four points throughout pregnancy and had health complications logged.

The researchers measured blood levels of nearly 3000 proteins and found one in particular associated with both preeclampsia and fetal growth restriction: ISM2, a little-studied protein produced by the placenta. Specifically, lower blood levels of the protein at the 12-week blood test were linked to higher risks of preeclampsia, fetal growth restriction, or both.

Checking the relationship in two other data sets—from a follow-up study called POPS2 and a separate study in Sweden—the researchers found the same pattern. What’s more, ISM2 levels were a better predictor of each condition than currently used proteins.

In lab experiments, the team found that blocking the production of ISM2 in stem cells that usually develop into normal, uterus-invading EVTs stopped that process from happening. By contrast, artificially adding ISM2 into a cell line that usually lacks it induced recipient cells to migrate around lab dishes more like EVTs would.

It’s clear “you need ISM2 to make an invasive EVT cell,” says study co-author Steve Charnock-Jones, a molecular biologist at Cambridge. The researchers don’t know how exactly ISM2 helps EVTs burrow into uterine tissue; perhaps it ups their sensitivity to other molecular uterine signals that promote invasion, such as cytokines, he says.

Sarosh Rana, a Gates-funded maternal-fetal medicine specialist at the University of Nebraska Medical Center who was not involved in the work, praises the researchers for going beyond merely associating ISM2 with disorders to reveal its likely role in them. “It’s good for people to do deep science—finding biological proteins that are relevant in actual biology,” she says.

However, she cautions that the study only compared ISM2 with other proteins in the blood, not with the full range of measures used to predict preeclampsia, so it’s hard to say how much of an advantage an ISM2 diagnostic would offer in locations where ultrasounds or other techniques are routine.

The research has already been snapped up by a San Francisco–based nonprofit company, Medicines360, for development into a blood test . (Charnock-Jones and colleagues are co-inventors on a patent describing the approach, and some are paid consultants for Medicines360.) The company is conducting a study to generate data for approval by the U.S. Food and Drug Administration and other regulators, notes study co-author Gordon Smith, a maternal-fetal medicine consultant at Cambridge’s Rosie Hospital.

There could be treatment possibilities, too. For women with too little of the protein, raising ISM2 levels might be beneficial—though technically challenging, Smith acknowledges. A more feasible therapy could be for ectopic pregnancy, a potentially life-threatening condition in which the embryo implants outside the uterine cavity. Blocking ISM2 with specially made antibodies could prevent EVTs from burrowing into tissue, starving the embryo of oxygen and nutrients and ending the pregnancy.

It remains unclear what causes some women to have less ISM2 than others, and this, along with understanding how the protein influences placental development, is a critical area for future research, Charnock-Jones says: “Why is there less? … And why does having less matter?”