Our A.I. and “Big-Tech” overlords are pleased the new AI browsers are here

One focus of mine as I dabble in AI tools, is to think about anonymizing my queries which, without privacy and security safeguards that I can easily implement, has turned out to be extremely difficult…and it’s getting harder.

The other afternoon I was talking with a friend in cybersecurity and he was ranting about all the ways third-party trackers (and obviously big-tech, governments, et al) are able to track us now:

1. Fingerprinting of browsers and devices.
2. Apps running in the background on a smartphone, tablet and yes, even a laptop/desktop that are able to triangulate one’s IP address (even a VPN one) with fingerprinting to “watch” traffic emanating from a specific device, even when that device is connected to a VPN. This is done by coordinating that analysis with connected apps phoning home simultaneously.

God forbid we now add extreme AI-driven behavioral and psychographic profiling to fingerprinting and IP-traffic analysis! Then having third-party tracking companies (or nation-states) deploying all of that data to an AI-based predictive model and you can see how explosively powerful it would be if one’s traffic was coupled with AI queries *and* browser traffic and our bookmarking!

It’s a fascist government’s and Palantir’s wet dream (read this Palantir article for more).

All I can say is OMG…it was hard enough to guard against privacy and security invasions but adding an AI web browser in to one’s browsing mix? An AI browser is a Pandora’s box of unknown threats!

So obviously I used A.I. to quickly ask what it thought. NOTE: I use DuckDuckGo’s Duck.ai to anonymize my queries and choose different models to see which is best and how they vary.

Hopefully this will help you think-through whether or not you  should even begin using and AI-browsers like Open AI’s Atlas or Perplexity’s Comet AI-centric browser.

——————————

MY QUERY TO CHATGPT5
“Are there any risks in downloading and using AI-centric web browsers such as OpenAI’s Atlas or the new Perplexity Comet browsers?”

CHATGPT’s RESPONSE
Yes. AI-centric browsers like OpenAI’s Atlas and Perplexity’s Comet introduce new risks beyond traditional browsers.

KEY RISKS

• Prompt injection: Hidden instructions on web pages can hijack the AI agent to exfiltrate data, mis-post on your accounts, or take unintended actions.

• Cloaked content manipulation: Sites can serve one view to humans and a different, AI-targeted version to mislead the agent (e.g., fake prices, reputational smears).

• Memory/data retention: Agent “memories” and broad context can store sensitive details across sessions; tainted prompts can persist until manually cleared.

• Over-privileged access: To work, agents may read emails, calendars, and dashboards and act on sites—expanding the blast radius if compromised.

• Extension/overlay spoofing: Malicious extensions or scripts can fake the AI sidebar, capture inputs, and inject commands.

• Immature defenses: Prompt injection is an unsolved problem; early products have had live vulnerabilities and rapid patch cycles.

• Traditional threats still apply: Malicious AI extensions, phishing, CSRF, token theft, and session abuse remain risks—and may be amplified by agent autonomy.

PRACTICAL SAFETY STEPS

• Limit access: Don’t connect email, banking, or admin accounts. Use read-only where possible. Disable autofill for sensitive sites.

• Segment usage: Use a separate profile or dedicated device for AI browsing; keep work and sensitive logins in a different browser.

• Constrain autonomy: Turn off agent auto-actions by default; require explicit confirmation for writes, purchases, posts, or credential use.

• Manage memory: Disable persistent memories or routinely review and clear them—especially after odd agent behavior.

• Harden extensions: Install minimal, well-vetted extensions only. Review permissions; remove anything you don’t need. Keep the browser updated.

• Zero trust with pages: Don’t let the agent summarize/act on logged-in dashboards, admin panels, webmail, or docs. Prefer static/public pages.

• Verify outputs: Cross-check prices, promotions, and critical facts directly on sites before acting. Treat agent summaries as drafts, not ground truth.

• Use strong session hygiene: Password manager, passkeys, MFA, per-site profiles, regular sign-outs from sensitive services.

• Network and OS hygiene: Keep OS/AV up to date; use DNS filtering or a reputable security suite; avoid installing beta builds on primary machines.

• Incident response: If the agent misbehaves, revoke connected app permissions, clear agent memory/history, rotate passwords, and review account activity.

Privacy. Get over it?

WHO SHOULD AVOID USING FOR NOW

• High-risk roles (finance, healthcare, legal, admin accounts), anyone handling regulated or confidential data, and users relying on extensions from unknown vendors.

BOTTOM LINE

• AI browsers can be useful, but today they carry materially higher security and data exposure risks than traditional browsers. Treat them as experimental, isolate them, minimize permissions, and keep a human-in-the-loop for any action that touches sensitive data or accounts.

SOURCES
• Venture Beat: https://venturebeat.com/ai/when-your-ai-browser-becomes-your-enemy-the-comet-security-disaster

• Hacker News: (Oct 28, 2025) https://thehackernews.com/2025/10/new-chatgpt-atlas-browser-exploit-lets.html

• TechCrunch: The glaring security risks with AI browser agents — https://techcrunch.com/2025/10/25/security-risks-ai-browser-agents-atlas-comet/

• Axios: OpenAI’s new browser sparks privacy, security concerns — https://www.axios.com/2025/10/28/openai-atlas-browser-privacy-security

• Fortune: Cybersecurity experts warn OpenAI’s ChatGPT Atlas is vulnerable to attacks that could turn it against a user — https://fortune.com/2025/10/23/openai-chatgpt-atlas-security-risks-prompt-injection/

• Columbia Journalism Review: How AI Browsers Sneak Past Blockers and Paywalls — https://www.cjr.org/analysis/ai-browsers-atlas-comet-paywalls.php

• CyberScoop: Exclusive: Researchers show how AI browser agents can be fed hidden, manipulated data — https://cyberscoop.com/openai-atlas-splx-research-cloaking-attacks-browser-agents/

——————————

What do you think? Should I just adhere to the premise made infamous by the once CEO of Sun Microsystems, Scott McNealy, who told a group of journalists, “You have zero privacy anyway. Get over it.” and just use these browsers anyway?

Where have I been? My wife, son and I moved from Minnesota to Southern California in 2018 (our daughter had been here since 2010) so we’re all out here and my bride and I are both officially retired these days. Buying and selling house, moving twice, wrapping stuff up and getting acclimated and poking around the desert southwest has meant a lot of adventures.

My objective with starting up again is to try and stop using other social media to journal about life, love, pursuit of happiness and connect dots once again. Hope you might find something of meaning as I’m going to try and add value to the world and not devolve in to ranting or negativity, but rather focus on the positives.

Enjoy.

My wife Michelle on our way to dinner via the London Underground and minutes from our mysterious encounter.

It was the summer of 1990 and my wife Michelle and I were in London for an adventure. It was a fun time and she had arranged for us to meet an old colleague of hers (and his partner) for a dinner at their townhouse one evening.

Michelle is a planner and really good at it. Me? Not so much, especially when on a vacation (I love to poke around and be spontaneous). That said, I felt like I should probably reach out to the one guy I knew in London, a fellow I’ll call Blake, who had gone there to open an office for a US-based brokerage company. I knew Blake, but not well. Still, he was a nice guy, we had mutual friends, so Michelle kept encouraging me to call him. I didn’t want to take time away from our own time together, even though I felt guilty the whole time for not reaching out to say hello.

We got on a London Underground subway train to head to her colleague’s townhouse which also required a short connection to a different line. As this first train stopped at our transfer destination, we saw the other one pulling in to the station across the way! “Let’s go!“, I cried out and we exited and raced up the stairs, across the footbridge, down the opposite stairs and she was pulling me toward one car but I said “No” and pulled her to a different car.

“Phew! We made it,” I exclaimed. the train car was packed with commuters. We got to the next station and people got off. Then the next station and people got off. Then another one. Now there were far fewer people on the train and, as I glanced up, standing three feet from me was Blake!

He looked at me, broke in to a grin, and said “Borsch! What are you doing here? Why didn’t you call me?” I stammered out a response, we chatted for a few minutes, and then we arrived at our stop. We said our goodbyes while all I could think was “How the f**k did THAT just happen?”

1987 map to the London Underground

Coincidence? Possibly, but I don’t believe in them.

Michelle was absolutely convinced (and is to this day) that it was my thinking about Blake that brought for this encounter. As I’ve tried to analyze this incident (and I’ve had several others in my life that are similiar, by the way) I think about these points and have so many questions:

• London’s population in 1990 was 6,775,000 but swelled to over 9 million during the day with people commuting in to the city. What are the chances that I’d have pulled Michelle in to the exact car—and at the correct end of that car—to be able to feet away from Blake?
• How could we have been traveling on the Underground at the exact same time to end up simultaneously being in that train car?
• What if I’d taken Michelle’s lead and gone on the other train car she was pulling me toward?

While I could go in to my own theories as a non-physicist (e.g., quantum entanglement means everyone we interact could have some level of entanglement with each of us) but that’s a discussion for another time. The mystery story is more fun and puzzling.

Be interested if something like this has happened to you. Comment below.

Since late 2022 it’s been nearly impossible to ignore all of the onrush of writeups, videos, and wild over-enthusiastic reactions to the major steps that have been made with artificial intelligence (AI), specificially OpenAI and its ChatGPT.

Like you have probably been doing, I’ve been staying aware of the baby steps, hyperbole and complete bullshit as it pertains to the fear AI will take over the world and destroy humanity as some pundits (and blowhards) intimate will occur.

Yes, cool things are happening with synthetic media (i.e., generative AI media) as well as text creation by ChatGPT-driven chatbots.

This is why I was pleased to see this discussion about “The Bing AI Overreaction” by Leo Laporte’s TWiT show Windows Weekly. Guests were long time Microsoft analysts Paul Thurrott and Richard Campbell. After watching this it *might* tone-down some of the over-enthusiasm and hyperbole of Bing’s “AI search” capabilities.

The laws that trivialize theft of this nature don’t factor in all of time, effort, energy and expense that a shipper goes through.

That said, imagine all of the people waiting for these packages? If there are a thousand package boxes lying on these tracks, think of the 1,000 people expecting them and adjusting their schedules to be at home?

But how much memory would I really need? I’ve always purchased as much RAM as I could afford, but these new M1-based CPUs are so efficient, I knew I could probably get by with less. Plus, the two Apple Geniuses I spoke with when in the store — who quizzed me about my needs at length — said I could easily get by with 16GBs of RAM and the Mac Pro version of the M1 CPU.

But I was not convinced…that is until I watched this video which, while not GeekBench like analysis, pushed me over the edge and convinced me to go with the 14″ model and only 16GBs of RAM…so I bought it that day:

I also picked up the LG 34WK95U ultrawide display along with a new OWC Thunderbolt 3 Dock and a fast Sandisk 4TB Extreme Pro SSD disk. I’ve got the perfect set up for what I need today, as well as for the next several years.

One more thing…

On Apple’s website they call these machines and their new M1 Pro CPU “scary fast” and the M1 Pro Max “scary faster.” I can attest to the fact that, now running this MacBook Pro 14″ in my own office with my apps and files on it, HOLY CRAP IS THIS THING FAST!

Photoshop used to open in about a minute. Now Photoshop opens in 6.5 seconds. It’s that way with all of my apps. Damn…is this thing ever fun to use.

So if you’re in the market for a new machine, I highly recommend it.

If you care about privacy and the security of your communications, Signal app is likely already on your phone, tablet or computer. If not, it should be and you should be using it.

Some time ago I became troubled that Cellebrite, an Israeli “digital forensics” tool, was law enforcement and government’s method of choice to extract data from iOS and other devices. Whether used for warranted legal investigations by law enforcement, or by oppressive governments looking to stifle dissent in their countries, the company once boasted that they could even hack Signal.

That was too much for Moxie Marlinspike and Brian Acton, the founders of Signal Foundation, who refuted Cellebrite’s claims in this blog post a few weeks ago.

Rather than go in to a lot of detail, let me instead point you to In epic hack, Signal developer turns the tables on forensics firm Cellebrite by Riana Pfefferkorn from the Center for Internet and Society at Stanford Law School. It recaps the entire controversy well.

Suffice to say that I do have concerns that Signal has poked the bear … and in this case the bear is Congress, the Justice Department, the National Security Agency, and any intelligence or law enforcement. I am certain that all of the above would rather have it continue to be trivial and simple to mass surveil the U.S. population and track us through our digital devices.

Perhaps this poking will accelerate U.S. laws being enacted to continue to erode end-to-end encrypted (E2EE) communications. This will make us all more susceptible to surveillance, but most importantly to hacking. Once a backdoor is put in, or some company finds a way in to gain access to our device’s data, hackers will quickly gain access too.

Yes, I do have concerns that E2EE makes law enforcement and intelligence gathering more difficult. But the tradeoff is literally making billions of people vulnerable to catch a fraction of them breaking the law. These agencies have significant tools in their arsenals and don’t need to make us all weaker to do their jobs.

While I’m the most benign person on the planet when it comes to secrets or having something to hide, there is NO WAY that I want my phone, tablet or computer to be MORE vulnerable rather than less. Same thing with my communications: What I say in a text message, voice or video call is no one else’s business over and above myself and the person(s) I’m communicating with at the time.

It’s finally published. This report is a stunning body of work and emphasizes the overwhelming need the United States has to be a (the?) dominant player in artificial intelligence (AI).

National Security Commission on Artificial Intelligence Final Report