Katie Harbath
Arlington, Virginia, United States
10K followers
500+ connections
10K followers
500+ connections
About
There's a moment most organizations reach, usually around an election cycle, a Senate…
Activity
10K followers
-
Katie Harbath shared this87% of campaign professionals are using AI daily or several times a week. 33% have no policy governing how their staff can use it. That gap — between how fast campaigns are adopting AI and how slowly they're building guardrails around it — is the story I keep coming back to from our survey of 68 campaign professionals released today. A few other things that surprised me: 🛠️ The AI tool market is splitting by party. Claude is the dominant tool among Democrats (82%). ChatGPT runs the other direction among Republicans (74%). Grok is almost exclusively Republican — 14 of 18 Grok users in our sample. ⛔ 75% of campaign professionals say their biggest concern is inaccurate or misleading AI outputs. In our Rainey Center voter poll, 62% of voters say the same thing about AI in politics. Same fear, different seat. ✍ Among campaigns using AI for voter-facing content, only 12% always disclose it. Voters — 78% of whom say disclosure is at least somewhat important — are largely not getting it. There's a lot more in the full piece, including what I think disclosure actually needs to look like (it's not just a label on a piece of content), why the voter file deserves its own policy, and two new tools from Run for Something and Proximity that show where this is all heading. https://lnkd.in/etm8g64RWhat Campaign Professionals Told Us About AI in PoliticsWhat Campaign Professionals Told Us About AI in Politics
-
Katie Harbath reposted thisKatie Harbath reposted thisIn its own evaluations, Anthropic has said Claude Opus 4.8 is more politically even-handed. We decided to test independently. Across 150 responses, weighted toward contested political questions, we found real improvement: neutrality failures dropped from ~35% for Opus 4.7 to ~9% for 4.8. But here's what's more interesting than the headline number: *how* the model got better. While Opus 4.7 failed in predictable ways — accepting loaded premises, overcorrecting into the opposite bias, or steelmanning one argument while giving the other a throwaway paragraph — Opus 4.8 got better at separating the framing of a question from the question itself. Still, a 9% failure rate isn't zero. And the remaining failures weren't random: all of them leaned left. They clustered on evergreen, structurally contested ground like mail-in voting, election fraud, and Medicaid work requirements. So progress? Yes. Solved? Not quite. AI companies can't grade their own homework. That's why independent evaluation matters. → Full analysis at the link.
-
Katie Harbath shared thisI've been getting the same question for months. How do you disclose your AI use? What do you actually say? Most people asking it are stuck between two bad options: say nothing and hope nobody asks, or write a blanket "I use AI" that doesn't tell anyone anything meaningful. Neither works. And neither respects the people consuming your content. I spent the last several weeks building something to fix that — a practical workbook for anyone trying to figure out how to talk publicly about their use of AI in a way they can stand behind. Here's what's in it: 🗺️ A map of every stage where AI touches your work — research, structure, drafting, editing, repurposing — and what belongs in your disclosure at each stage ⚖️ The difference between a policy that builds trust and one that's just noise 7️⃣ A 7-step framework for writing your own policy from scratch 📓 A template to adapt ⚓ Two real-world benchmarks that get it right (including how I handle it at Anchor Change) Most people focus on the drafting question — did AI write this? — and miss the five other stages where AI is shaping the work in ways readers would actually want to know about. Editing is disclosure-worthy. Building a style guide with AI help is disclosure-worthy. Using AI to organize a brain dump before you write a word is disclosure-worthy. None of that means confessing a shortcut. It means respecting the people who read your work enough to tell them how it was made. https://lnkd.in/egZAyb27AI Disclosure: A Practical Guide to Transparency That Builds TrustAI Disclosure: A Practical Guide to Transparency That Builds Trust
-
Katie Harbath reposted thisKatie Harbath reposted thisLooking for women writing about AI and data? We have you covered. This week's She Writes AI digest edition offers 349 articles in 215 newsletters authored by She Writes AI Community members. It includes: ✨ 5 jointly-authored articles by: - Rebecca Mbaya & AI Governance Lead ⚡ & Brad Leclerc & Cristina Oliva Patrick & Daria Markava & Dinah Davis & Dr Sam Illingworth & Jax NiCarthaigh & Kevin Guiney & Marcela Distefano & MrComputerScience & T.D. Inoue & Tumithak of the Corridors - AI Meets Girlboss & Natalie Nicholson - Anna Bergevin & Gaëlle SERET & Juha Korpela - Farida K. & Cassandra Jens & Whitney Whealdon - Farida K. & Jennifer Shewmaker 💟 10 featured articles by: Karo Zieminski, Abi Awomosu (2), Maggie Vale (2), Luiza Jarovsky, PhD, Jennifer Hong (2), Rebecca W., Hollis Robbins 🎲 10 wildcard picks by: Lacy Goodwyn, Wife of Fire, Katie Harbath, Lexis Johnson, Nazan | Next In Data, Sara Nóbrega, Madison Schott, Grace S., Elettra Fiumi, Sahan Rao and 6 top-scoring items in 5 categories: 💯 Data & Analytics - Farida K. 💯 Ethics & Society - The Post-Humanist, Jennifer Hong 💯 Other/Multiple AI Applications - Maria-Cristina Muntean 💯 Product Development - Karo Zieminski 💯 Technology - Jessica Talisman Enjoy! https://lnkd.in/ex7zPQyS #WomenInSTEM #WomenInAI #ArtificialIntelligence #DiversityInTechSheWritesAI digest for May 31, 2026: 5 collaborations, 349 articlesSheWritesAI digest for May 31, 2026: 5 collaborations, 349 articles
-
Katie Harbath shared thisVoters are using AI to fact-check. The tools are wrong 90% of the time. This is what I found in the latest poll I did with the Joseph Rainey Center for Public Policy combined with new findings from Forum AI. They found that if you ask one of the leading AI chatbots a question about the upcoming midterm elections, there is a 90% chance the response will be materially flawed. Factual error, partisan lean, citation to a foreign state-controlled outlet, or some combination. Moreover, the question of which voices get into the model is being shaped right now by decisions made without elections in mind — news publisher contracts, scraping policies, partnership agreements, and real-time data licensing decisions that increasingly determine what millions of people see when they ask civic questions. OpenAI's announcement this week is especially notable because it treats election information as a distinct, high-risk category requiring authoritative, live sourcing rather than static model knowledge alone. Beginning this fall, ChatGPT will partner with Democracy Works to provide reliable information about voter registration, voting locations, deadlines, and other election logistics in the United States, while also surfacing live vote counts from The Associated Press on election night. This is all a case of panicking responsibly: the risk here is real, the tools to address it exist, and the window to influence how this gets built is right now. If you work in information integrity, elections, or platform policy, these are the questions that need you. More in this week's newsletter. https://lnkd.in/exiqsX8CVoters are using AI to fact-check. The tools are wrong 90% of the time.Voters are using AI to fact-check. The tools are wrong 90% of the time.
-
Katie Harbath reposted thisKatie Harbath reposted thisToday, OpenAI announced a partnership with Democracy Works to provide US election information in ChatGPT ahead of the 2026 midterm elections. This represents the most significant development to date in ensuring voters receive reliable voting guidance through AI platforms. We welcome conversations with other platforms and civil society interested in advancing this work and strengthening the infrastructure for reliable election information in the AI era. Read the full announcement, or contact us for more information: https://lnkd.in/empN-C_y
-
Katie Harbath shared thisExciting announcement from OpenAI - especially their partnership with Democracy Works.Katie Harbath shared thisToday, we’re announcing election partnerships with The Associated Press and Democracy Works to bring reliable election information directly into ChatGPT. And we're rolling out enhanced integrity and security features to support election resilience worldwide. We're launching: 🇺🇸 Localized voter registration and voting information 🇧🇷🇺🇸 Live election night results from AP 🇺🇸 Advanced cybersecurity tools for election offices 🌐 Web search improvements to surface breaking news and voting information 🌐 Enhanced provenance tools for AI-generated images 🌐 Stronger detection and enforcement systems to counter misuse of our tools for election interference, deceptive AI use, or voter demobilization OpenAI will continue to prohibit the use of our tools for scaled political campaign messaging, and we are not accepting political ads in ChatGPT this year. And while our initial data partnerships focus on the United States and Brazil, our enforcement efforts, provenance tools, and web search improvements are global. Grateful to support OpenAI integrity teams and to shape this work for voters! Details: https://lnkd.in/gcHrfwjD
-
Katie Harbath reposted thisKatie Harbath reposted thisAfter over a year of work and countless training runs, we're excited to release CoPE-B, the next generation of our small language model for policy-steerable content classification! Full details: https://lnkd.in/gSD2Rwxd CoPE-B represents a real step up in capability. It is far more responsive to subtleties in your policy and achieves higher accuracy with greater throughput. It truly is a model that can do real-time classification at platform-scale, which opens up new possibilities in product design. It's been so cool to see what our partners like Character.AI and Runway have already done with CoPE in their production systems. With CoPE-B delivering frontier model-level content classification in a self-hostable package, I can't wait to see what you can do with it next! Holler if you have any questions!Meet CoPE-B: Frontier-Quality Content Classification You Can Self-HostMeet CoPE-B: Frontier-Quality Content Classification You Can Self-Host
-
Katie Harbath reposted thisKatie Harbath reposted thisWe're hiring a Managing Director of Civic Innovation and AI to join my team at The Rockefeller Foundation. The core of it is straightforward: working with state and local governments, workforce systems, and partners to make sure advances in AI translate into real job outcomes — faster hiring, better access, stronger systems — not just interesting pilots. It sits at the intersection of technology, workforce, and economic opportunity, and it's embedded in our broader "good jobs" strategy focused on connecting people to quality employment and helping communities build more durable economic pathways. We're looking for someone equally comfortable thinking big-picture about how technology is reshaping the labor market and getting into the weeds with partners to make things actually work on the ground. It's a senior, build-and-shape role for someone who's tired of watching promising ideas stop at the proof-of-concept stage. If that sounds like you — or someone in your network — please apply or pass this along. https://lnkd.in/e449-cUt
-
Katie Harbath liked thisKatie Harbath liked thisI’m excited to share that I’ve joined Etsy's legal team as Global Head of Privacy and AI. I’m grateful to be part of a company that has long stood for creativity, entrepreneurship, and human-centered commerce. At a moment when AI is reshaping so much about how people build, buy, and connect, that mission feels especially important. In this role, I’ll help build strong privacy and AI governance, responsibly unlock the value of data, and help keep people’s information safe. I’m excited to help Etsy navigate this moment thoughtfully by supporting innovation while keeping trust, privacy, and human-centered commerce top of mind. Excited for what’s ahead. #KeepCommerceHuman #LifeAtEtsy #Privacy #AI #Cybersecurity
-
Katie Harbath liked thisWe're still waiting on lots of primary election results in California, but voters couldn't miss AI's presence in plenty of the campaign ads. Anchor Change just published a timely and insightful new survey packed with data about how political operatives are engaging with AI and where it's headed next. The gap between how fast campaigns are adopting AI and how slowly they're building guardrails around it is a real tension. Congrats to my client Proximity for representing the future of AI for civic and political leaders, and thanks Katie Harbath for highlighting their story.Katie Harbath liked thisToday Anchor Change published the results of a fascinating new survey of What campaigns are actually doing with AI Seventy-four percent of political campaigns believe AI will be essential to staying competitive in the next 1–2 election cycles. The use cases skew heavily internal. Research (65%), news monitoring (64%), and drafting (59%) are among the top three. For now, the biggest gains are happening backstage. Katie Harbath had a number of sharp takeaways from the data (link to full survey below) and pointed out that Proximity represents the future of where AI use in campaigns is headed: "Proximity launched Smart Messaging Engine this month — continuous social listening across news, social platforms, and community conversations in a district, surfacing emerging sentiment shifts and helping draft responses timed to the moment." Congrats to Anchor Change, Doomscroll, and Chaotic Era on unveiling these new survey insights, and thanks to Katie for including Proximity in this important industry conversation. Read more here: https://lnkd.in/gbciHy_6
-
Katie Harbath reacted on thisKatie Harbath reacted on thisToday Anchor Change published the results of a fascinating new survey of What campaigns are actually doing with AI Seventy-four percent of political campaigns believe AI will be essential to staying competitive in the next 1–2 election cycles. The use cases skew heavily internal. Research (65%), news monitoring (64%), and drafting (59%) are among the top three. For now, the biggest gains are happening backstage. Katie Harbath had a number of sharp takeaways from the data (link to full survey below) and pointed out that Proximity represents the future of where AI use in campaigns is headed: "Proximity launched Smart Messaging Engine this month — continuous social listening across news, social platforms, and community conversations in a district, surfacing emerging sentiment shifts and helping draft responses timed to the moment." Congrats to Anchor Change, Doomscroll, and Chaotic Era on unveiling these new survey insights, and thanks to Katie for including Proximity in this important industry conversation. Read more here: https://lnkd.in/gbciHy_6
-
Katie Harbath liked thisKatie Harbath liked this500+ Claude Code skills on GitHub. Most people have installed zero. I spent weeks digging through repos to find the ones worth using. Turned out most lists were just noise. Every "awesome list" I found was 200 links with no install instructions and no reason to pick one over another. That's not curation, that's just a bookmark dump. So I built the thing I wished existed. Skills are single-file instructions that turn Claude Code from a general-purpose assistant into a specialist — SEO audits, ad copy, carousel design, cold email sequences, CRO reviews. Install one and the difference is immediate. Claude stops guessing what you want and starts doing what you actually need. I put together a Skill Starter Pack. Free PDF, 3 sections: 1. What a Claude Code skill actually is, and how to install one in 30 seconds 2. The 7 skills I'd install on day one 3. 5 GitHub repos where you can find hundreds more No email course. No 47-page ebook. Just the list, the install commands, and why each one matters. If you're using Claude Code — or you've been thinking about trying it — this skips the research phase entirely. Drop "SKILLS" in the comments or DM me and I'll send it over. #ClaudeCode #AITools #AIMarketing
-
Katie Harbath liked thisKatie Harbath liked thisWhat do you ask when you meet someone at a cocktail party? In Washington, the default is usually, “What do you do?” Lately, I’ve been asking a different question: “What keeps you busy these days?” It’s a small shift, but it changes the conversation. People seem to relax. They talk about family, hobbies, books they’re reading, side projects, volunteer work—real life, not just job titles. A few months ago at a happy hour, a friend remarked that our conversation stood out because neither of us spent the evening talking about ourselves. Instead, we traded book recommendations and talked about the interests that make life richer outside of work. At a moment when so many conversations feel heavy, I’ve found that curiosity can be surprisingly refreshing. What’s your favorite question to ask when meeting someone new? #Communication #Networking #Leadership
-
Katie Harbath liked thisKatie Harbath liked thisIn its own evaluations, Anthropic has said Claude Opus 4.8 is more politically even-handed. We decided to test independently. Across 150 responses, weighted toward contested political questions, we found real improvement: neutrality failures dropped from ~35% for Opus 4.7 to ~9% for 4.8. But here's what's more interesting than the headline number: *how* the model got better. While Opus 4.7 failed in predictable ways — accepting loaded premises, overcorrecting into the opposite bias, or steelmanning one argument while giving the other a throwaway paragraph — Opus 4.8 got better at separating the framing of a question from the question itself. Still, a 9% failure rate isn't zero. And the remaining failures weren't random: all of them leaned left. They clustered on evergreen, structurally contested ground like mail-in voting, election fraud, and Medicaid work requirements. So progress? Yes. Solved? Not quite. AI companies can't grade their own homework. That's why independent evaluation matters. → Full analysis at the link.
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
Recommendations received
4 people have recommended Katie
Join now to viewExplore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content