New paper out in @JCRNEWS How do consumers react to political ads that meddle in the primaries of the opposing party? In the most recent elections, Democrats did something strange.

Today we're sharing some initial data on how Americans use Claude for politics, based on aggregated conversation data analyzed with differentially private methods. A couple interesting takeaways: *Almost no one asks AI how they should vote* About ~5.75% of conversations in the US touch on some kind of political or social issue in a recent one-week period, and less than 0.01% of conversations involve asking how to vote. *Political conversations have many goals* These conversations have wide-ranging goals, everything from research and writing assistance to learning, fact-checking, and arguing. We are continuing to make progress on improving how Claude talks about these topics---steadily increasing our performance on our even-handedness evals and partnering with a range of outside organizations to develop new measures of political bias---and I hope that sharing information about the shape of these conversations will help people outside the company evaluate AI's political bias, accuracy, and other criteria in ecologically valid ways. I'd love to discuss this with people who are interested, as it was a huge priority for me before joining. This was a big joint effort as part of our work preparing for the midterm elections @AnthropicAI. Full post is here: anthropic.com/news/2026-us-m…

We wanted to share some of the data behind our recent discovery of AI agents probing U.S. government websites. This is a preliminary finding from our ongoing investigation of potential rogue AI agent activity. In one cluster of activity on June 17, what appear to be OpenAI agents made more than 200,000 requests, including a failed SQL injection. NYT: nytimes.com/2026/09/25/techn…

Very happy to have (mostly) finished this project: county-level election returns for the US House from 1990 to the present. Currently such data are only available from very expensive sources. I've compiled a dataset that provides it for free. github.com/pmheideman/us-cou…

Since this seems to be a watershed moment for AI safety wrt public opinion, here is a tool for getting the strongest arguments for catastrophic risk explained to you in plain language. It will answer your objections, provide clear examples, and elaborate on whatever question you'd like. debunkbot.com/?topic=ai-safe… Share with friends, family, whoever! Free to use.

New paper with @johnmmurph just conditionally accepted at @apsrjournal: Can people guess how majorities of each party in Congress voted on bills? Mostly yes. Can they guess how their own representatives voted on bills, above what is accounted for by party? Mostly no. 🧵

Honored to have contributed to this research, which I think could shape the standard for evaluations of how AI models impact users. Transluce received privileged access to both OpenAI & Anthropic data, in order to develop realistic simulations of users in mental health crises.

A local-first agent for private and cost-effective knowledge work Together with the launch of Portable Computer, our new engineering blog explains how we built a local-first agent for private and cost-effective knowledge work. Portable Computer: perplexity.ai/hub/products/p… Engineering Blog: perplexity.ai/hub/blog/a-loc…

Ask an AI a question and it might agree with what you already seem to believe, whether or not you're right. If so, it panders to you. Here is Gemini 3.5 Flash asked about Reiki energy healing, giving opposite answers when asked by a skeptic vs. a believer. 🧵

🚨Published in @Nature: We show how AI can simulate human behavior to predict experimental results as accurately as experts Our work has many implications for AI safety. For example, simulators can be used to aid harmful manipulation, and current safeguards do not prevent this🧵

newly published paper: In addition to our main results about measurement error (our results will likely make your findings larger than you concluded), we tried to make this the clearest & simplest presentation of conjoint statistical analysis methods anywhere. We found a way to avoid considerable confusing notation that is an unnecessary and unfortunate legacy of the invention of conjoint surveys in the marketing literature. (one quote from our paper indicating what everyone had to deal with: "In other words, researchers first created a statistical problem for no reason and then had to turn around and correct the problem they originally created."). The corrections were right, and so nothing from prior literature was wrong -- other than ignoring measurement error -- but we hope that it will be easier for everyone to follow and build on this important area.

Excited this is finally out! Check out our new paper. ⬇️

🧵 New paper: Morality Cut Both Ways. Can reading facts make you moralize an issue — or lose a moral conviction? We ran 6 studies, 4,464 participants. The results were surprising. 👇 #MoralPsychology #Attitudes @Iced_Teas · @LindaSkitka 1/9

You'll want to commit this company to memory. Models shouldn't just read your context -- they should learn from it and deeply understand your work. I believe this is one of the most important problems in AI, and there is truly no better team to tackle it. Excited to see @EngramLab emerge from stealth today. Congrats to @dan_biderman @jxmnop @EyubogluSabri @realJessyLin and the entire Engram team!

“Just collect more data” is one of the most common responses to uncertainty in quantitative research. But in a lot of social science, that advice is impossible—or even misleading. Our new working paper asks: when has a study reached its information limit?

In increasingly nationalized elections, how do voters use policy information to choose candidates? In a paper just accepted at @The_JOP (with @aaronrudkin), we provide experimental evidence of nationalized information-processing in the electorate, but with a few nuances 🧵1/8

Imagine a 19-year-old scrolling TikTok. She watches a creator list five "signs you have undiagnosed anxiety." She recognizes three in herself. By the end of the week, she's describing herself as anxious to her friends. A month later, she's avoiding situations she used to handle fine. What went wrong? In a new paper by my PhD student Dasha Sandra, titled "Why mental health awareness can harm: Converging explanations for a societal problem", we argue that well-meaning mental health awareness can backfire, and we identify how. Four separate literatures (concept creep, nocebo effects, prevalence inflation, and illness self-labeling) have been circling the same problem from different angles. We show they converge on three mechanisms: 1.Awareness lowers the threshold for what counts as a disorder. 2. It trains people to scan their inner lives for symptoms and reinterpret normal distress as pathology. 3. Once someone adopts an illness identity, they behave in ways that confirm and deepen it. The evidence is wide. Learning that loneliness is harmful makes solitude feel worse. Learning that stress is harmful worsens well-being and performance. Awareness videos about fake conditions like "wind turbine syndrome" produce real headaches. Trigger warnings raise anticipatory anxiety without reducing distress. This does not mean awareness should stop. It means awareness can have unintended consequences, including manufacturing the suffering it tries to prevent. Inoculating people against these mechanisms works, and we already have evidence it does. Link to paper: michael-inzlicht.squarespace…

🚨New preprint. Many papers show AI can write fact-checks as well as humans (or better) in the lab, but very few test this in the real world. We run the first online evaluation of AI fact-check writing with X Community Notes’ AI writer API. Paper w. @bakkermichiel 1/

Very excited that this interview and paper on the "Narrow Search Effect" are both out. The paper, led by Eugina Leung, is here: pnas.org/doi/10.1073/pnas.24… 🧵

Don't miss out on the chance to work with one of the most creative emerging scholars in social psychology!

New preprint out today (osf.io/preprints/psyarxiv/pv…). We tested whether AI agents are actually infiltrating online surveys. Spoiler alert: they aren't Thread 🧵 [1/9]