Turning the OpenAI model off is possible. Turning off the open weights model distilled from the OpenAI model is not. I'm glad they're doing these evals and really hope we see more of this on training checkpoints, not just at deployment.

One intuition behind many AI policy proposals is that downstream AI applications -- the companies deploying AI in healthcare, finance, education, customer service, etc. -- should bear responsibility for ensuring safety. Our paper asks: What incentives does that create for the firms building AI?

“If recommendation algorithms gave online platforms the ability to control what people see online, agentic AI offers them an opportunity to control what people do.” Rutgers law professor Ellen P. Goodman considers Sen. Mark Warner's draft AI AGENT Act and details obstacles to its implementation.

Everybody calls this a salty sombrero, right?

Yet another departure from the safety team at OpenAI. www.businessinsider.com/openai-safet...

If current trajectories of AI development continue, it is highly plausible that AI will drastically transform our economies. We must make collective, democratic choices, rather than letting market forces play out and risking leaving most citizens behind. digitaleconomy.stanford.edu/news/wemusta...

@ljsmart.bsky.social Hi infodiva! Long time no see!

Noam Brown (researcher @ openai) He points out that performance improves consistently with reasoning time This explains a lot of what looks like benchmaxxing — performance isn’t plotted against cost It also has some serious implications for safety x.com/polynoamial/...

There's an OMB rule change that would require political oversight of scientific grants, and could pull any that don't "reflect presidential priorities" Another country has tried this before! It was the USSR and it was called Lysenkoism. By @thsaey.bsky.social www.sciencenews.org/article/omb-...

Trying to build an agentic skill in codex/claude code reminds me of the old Asimov story "Profession". It takes a ton of skill to build a top notch agentic skill! A mix of domain expertise and technical expertise in coding, agentic & context engineering. I suspect currently is very much an art

I will literally follow any and all ML PhD students on this website, where you at

AI appears to be finding software vulnerabilities at scale. In June 2026, 21 notable organizations disclosed ~1,500 high- and critical-severity CVEs, over 3.5× the previous monthly record set before Claude Mythos Preview's release.

The paperback of “Invisible Rulers: The Ppl Who Turn Lies Into Reality” comes out in 1 mo. It has the story of the machine that went after us in a new epilogue.(Invisiblerulers.com) Media folks, pls cover this loss for AFL. I’m very happy to talk ab it. The public needs to understand these attacks.

Today we beat Stephen Miller’s lawsuit mill in court! For 3 years I’ve been a defendant in a case filed by America First Legal on behalf of Gateway Pundit & a random antivaxxer we’d never heard of until she sued us. The plaintiffs claimed we “censored” them. It was a lie. They just lost.

WATCH: Chief Science Officer Dr. Jedidah Isler’s full opening statement before Congress on scientists as the key to American competitiveness www.youtube.com/watch?v=b6IFljG8QQM

"public trust in science and scientists is high...The idea that there’s a pervasive lack of trust is completely unfounded...But trust has dropped in certain groups...politically right-leaning groups" Maybe it's not science/scientists that's to blame... 1/2 www.nature.com/articles/d41...

Researchers discovered that people found AI impersonators to be more authentic, coherent, and relevant than the real politicians, raising alarm bells around the potential for public deception.

Neat look at how America celebrated previous anniversaries via @archive.org blog.archive.org/2026/06/29/t...

The Dead Salmons of AI Interpretability arxiv.org/abs/2512.18792 The famous #fMRI cautionary tale re-born as an analogy for the AI age! @practicalfmri.bsky.social

In a new interview with the Washington Post's Benjamin Guggenheim, ControlAI's US Director Connor Leahy explains why nobody really understands how AI works, the race to uncontrollable superintelligence, and what he's telling Congress. www.washingtonpost.c...

The people who were screaming the loudest about freedom of speech have instituted a banned words list. "Every accusation is a confession" continues to describe this administration well and raises serious concerns about their voter fraud accusations.

Dimensionality reduction is how the brain acquires knowledge. Cool new paper from the late, great, Mark Stokes. www.nature.com/articles/s41... #neuroscience

Criticize this all you want, but it's also worth asking yourself "what have I done?"

Sanders releases trove of internal HHS emails showing RFK Jr. pressured CDC over vaccine messaging Emails indicate that Kennedy directed the CDC’s vaccine advisory panel to restrict access to vaccines thehill.com/policy/healt...

If you're an #NSF funded researcher and haven't written a response to the OMB proposed rule and contacted your Congressional delegation about it, you should know Russ Vought is coming for your field first, enshrining the chaos of DOGE as policy. Do both at fight2win.standupforscience.net

AI researcher Nate Soares on Fox News: "It could be that next summer AIs can make a smarter AI that makes a smarter AI... and if that's true, it could be that in five years we're all dead." Source: Will Cain Country

But the elephant being ignored is Anthropic itself. It, and other frontier labs, are racing towards superhuman AI, dragging Chinese copycats along for the ride. The U.S. should take steps to make it harder for China to access dangerous capabilities, but this is not enough;

What else would you expect from a "hospitality" company whose goal is to automate everything & minimize any contact with guests.

My editorial in this week’s Science @aaas.org www.science.org/doi/10.1126/... 🧪

Anthropic's Mythos AI reportedly broke into almost all of the NSA's classified systems in hours. This week, Five Eyes agencies warn we must act now. What's happening, and why AI cyberhacking is just the tip of the iceberg:

Yesterday Alex Bores lost the NY-12 primary. But in a way, everyone won

AI is not just coming for your job. This PauseAI campaign has brought together hundreds of concerned citizens across the globe, building a movement to implement a pause in the development of frontier AI systems. If you are worried about the impact of AI, join us.

Using AI to support peer review seems unavoidable, but what quality checks should AI implement? We can take some lessons from metascience on the hard reward design problem that is AI review. I wrote a paper synthesizing a few points the emerging lit seems at risk of confusing. 1/

AI has mastered the power of persuasion. Are we really ready for machines to impose their intelligence on us? lnkd.in/eumGxW6z

Scientific American is being sold to LabX Media, publisher of Discover: group.springernature.com/gp/group/med... They've fired me and other union staff, ahead of June 26 vote ratification. My rates are suddenly more reasonable, if you need editing or reporting - [email protected]

🚨 #Evol2026 attendees - our team at @arcadiascience.com is hiring! We're looking to hire a computational evolutionary biologist interested in asking how and why protein sequences/structures/functions differ across the tree of life 🧬 jobs.lever.co/arcadiascien... Please share widely!

I have been watching the dialogue on this one closely. To everyone who is saying "OP is special, no one could use AI this way" well, I have spent 18 months doing significant amounts of scicomm in patient communities and they are ALL using AI this way. wrote this to help: github.com/DrCatHicks/i...

'USAID was actually widely recognized as one of the most efficient agencies across the government. As I said, 92 million lives it over the last two decades alone on less than one percent of the federal budget.'

Don't want to listen to the US military on AI risk? Fine, how about Australia, Canada, the UK, and NZ? www.theguardian.com/technology/2...

Dispiriting, candid look at the “civic education” movement which makes me feel that the real crisis of the humanities rn is utter collapse of belief that they can produce new knowledge—replaced (on the right and sometimes the left) by a belief that they are morally or politically good for you

1/ LLMs learn narrative from their pretraining data but what narrative content is actually in there? It turns out narrative is wildly unevenly distributed across sources and topics. New preprint with @andrewpiper.bsky.social @elliottash.bsky.social @mariaa.bsky.social:

I'm not saying I'm like one of those old people who read the obituaries, but after I read the one about Major Oak, I had to read about Oliver Tree. I thought the tree was a person and the person was a tree, but it turns out both were very different human tragedies. en.wikipedia.org/wiki/Portal:...

Pew Research published several new polls today on how Americans view AI. Several results struck me as interesting… Read the results and Joe’s opinion on them:

Big news: after we demanded they stand down, the Trump admin stopped its plans to rip out 80% of our nation's ocean research equipment. This proposal should never have seen the light of day. Our coastal economies deserve better than being treated as collateral damage in an ideological crusade.

Conclusion agentic search is coming for library/academic search and is going to be crazily disruptive

Stuart Russel sharing hopes that we don't need a huge disaster to do what we already know we should do re: AI: www.theguardian.com/commentisfre...

Powerful piece from two former NIH staffers www.insidehighered.com/opinion/view...

Schneier on opening the Pandora's Box of AI capabilities: www.theguardian.com/commentisfre...

I’ve seen the paper. It’s not a jailbreak. It was Defense Oriented Prompting (DOP) - a capability defenders need. My thoughts about the hasty Export Controls that made Anthropic halt access to Fable. If national defense is the goal, this is an own goal against us www.wsj.com/tech/ai/anth...