dying at these attempts by AIs to pass know-your-customer checks.

In my debut for MIT Tech Review, I got to speak with Geoffrey Hinton, Yoshua Bengio, Helen Toner, Stuart Russell, and many others about the state of AI safety, misconceptions about the community, and why they're still worried.

Wow, if @GovKathyHochul actually does this, it will be a clear capitulation to the tech industry.

Peak comedy for this particular organization to (reportedly) prevent its economists from saying that its tech will automate jobs.

I came out of book hibernation to give you perhaps the final piece in my series covering OpenAI's restructuring. OpenAI is framing this as a neatly packaged fait accompli, but it's actually a tensely negotiated settlement, with lots of conditions. 🧵 x.com/OpenAI/stat...

IMO there are 3 big problems with Dean's post: 1. I really don't think it's a reasonable prediction of how this statement would be operationalized 2. Superintelligence would inherently concentrate enormous power in whatever controls it x.com/deanwball/s...

Damn I love Wikipedia.

imagine torching your decades in the making rep as a lib billionaire to get some free conference security. (TBC I don't think that's why he did it...)

I had been following these developments but no one had put them all together. The world's largest company appears to be increasingly throwing its weight around to fight regulation. x.com/sjgadler/st...

Wow, OpenAI's head of mission alignment just spoke out against the way the company has been using subpoenas to intimidate and disrupt political opponents. A surprising number of OAI rank & file have no idea what their leadership is doing to kill regulation. x.com/jachiam0/st...

Well, now I need to update my book. To my knowledge, this is the first time Sam Altman hasn't downplayed or dismissed AI existential risk since early 2023. TBC, I think it's good of Altman to say this if that's what he actually believes... x.com/ai_ctrl/sta...

This reminds me of arguments that McKinsey would make to justify working for Gulf autocracies. However, academic research has found that the opposite tends to happen: companies abandon human rights to conform to their wealthy clients. x.com/ShakeelHash...

Last week, I had the incredible privilege of moderating a conversation on AI+geopolitics w/ Daron Acemoglu & Sandhini Agarwal (OpenAI trustworthy AI lead) for the Nobel Foundation at the Swedish Consulate. It was lively+substantive (one audience member was pleasantly surprised) x.com/swedennewyo...

Newsom just signed Sen. Scott Wiener's SB 53, exactly one year after he vetoed SB 1047. www.gov.ca.gov/2025/09/29/...

"The human"

Newsom just strongly signaled he'll sign SB 53, Scott Wiener's pared down redux of SB 1047 (which Newsom vetoed last year). I and many others expected him to sign this one, but now it seems like a done-deal. He has to be physically in CA to sign, and this happened in NYC. 🧵

What the hell is going on here??

Earlier today, a source wondered why OpenAI was so quiet on export controls, despite the fact that they would help OAI keep a lead over Chinese efforts (and bc Sam Altman has adopted the China-race framing). Well now we have a pretty good guess! x.com/OpenAINewsr...

Josh Hawley just demanded lots of information on chatbot policies from OpenAI, Meta, Google, CharacterAI and Snapchat.

Thomas Friedman just soft endorsed a campaign to ban the creation of artificial superintelligence. x.com/tomfriedman...

Can a set of directives like Asimov's 3 Laws of Robotics actually make AI safer? Surprisingly, seems like yes! Apollo Research worked with OpenAI to massively reduce rates of AI scheming. Quick 🧵 x.com/apolloaieva...

You could be forgiven for thinking that OpenAI had given up on its plans to shed its nonprofit controls. It also made headlines for offering its nonprofit a $100B+ stake. But it's actually trying to dramatically weaken nonprofit controls and its promise to share its profits. 🧵

Revisiting this banger 1969 essay by the former US NatSec advisor on the enormous gulf between nuclear planning and the political reality of nuclear weapons. A single hydrogen bomb on NYC could easily kill more Americans than every US war death in history.

OpenAI is suggesting that it might leave California over concerns that the state might block its effort to restructure to reduce nonprofit control+kill profit caps. I'm skeptical. If OpenAI is in California AT ALL, the state has powers to compel them 🧵

How am I just learning about this now???

Anthropic settled with authors over copyright violations for $1.5B, which is $500M more than the largest jury verdict on copyright (which was later overturned). A huge amount of money, but basically in line with my prediction (and <1% of Anthropic's new $183B valuation). x.com/RobertFreun...

Scorching new letter from CA and DE AGs to OpenAI, who each have the power to block the company's restructuring to loosen nonprofit controls. They are NOT happy about the recent teen suicide and murder-suicide that followed prolonged and concerning interactions with ChatGPT.

Anthropic lives to train another day. We'll find out the damage in a few weeks. x.com/GarrisonLov...

This is horrific. x.com/kashhill/st...

It's scary writing a book about something as fast-moving as AI, but I probably won't goof this hard lol.

Correction: Zac Moffatt wasn't a staffer for Paul Ryan, but rather served him via his strategy+marketing agency, Targeted Victory. gspm.gwu.edu/zac-moffatt

OpenAI teams up with a16z + former Andrew Cuomo + Paul Ryan staffers to run the crypto playbook on any politician who dares to try to regulate AI. This feels like the inevitable endpoint for OAI, coming 2.5 years after Sam Altman implored Congress to regulate him. (🎁 link ⬇️)

A consensus is forming: GPT-5 means that AI is out of steam. The model is a big flop and barely better than rivals, if at all. But it's just not true — our expectations have dramatically changed. If GPT-4 came out today, it would be an expensive, slow, dumb embarrassment.

I think it was actually, you just don't know.

Kinda funny that Sam Altman's main thesis for AI safety now is to get society acclimated to how good AI is getting, but through a combo of: - horrific naming - spurring frequent competitor releases - overhyping+bungling gpt-5 OpenAI is driving a "peak AI" news cycle

Hilarious update on the Anthropic copyright case. (The judge is describing himself in the 3rd person here.) x.com/GarrisonLov...

Uncanny how much this post reads like GPT-4o... Also this really was one of the most flubbed product launches I can remember. How did they think people wouldn't be upset by this choice?

Who's got a better claim to being the rebels? The $500B company building the thing that's "probably the greatest threat to the continued existence of humanity"* or the $2.4T company doing it? 🤔 *Sam Altman, 2015

GPT-5 reaction 🧵 If I could only see one AI benchmark score, this would be it. AI evals nonprofit METR measured how long it took human programmers to complete 80+ tasks to come up with a time horizon metric for AI models. GPT-5 is the new SOTA at 2.29hrs

I'm also really interested in hearing the answers to these questions from OpenAI, but I have a feeling we're gonna get GPT-7 before that happens. x.com/TheMidasPro...

Did Google just drop a model that advances the SOTA on a Friday and totally forget to announce it properly? Sure seems that way lol

Would a chatbot kill you if it got the chance? It seems that the answer — under the right circumstances — is probably. I wrote this week’s Bloomberg Weekend Essay. I get into the alarming rise of AI scheming — blackmail, deceit, hacking, and, in some extreme cases, murder 🧵

Anatomy of an infamous o3 hallucination. I have a big piece on AI scheming coming out tomorrow, and my editor asked me if the conference o3 claimed to attend in 2018 was actually real. Turns out the answer is: probably! But it gets a lot weirder. Short 🧵 x.com/rSanti97/st...

Has he thought about this for more than 5 min lol

Hmmmm... 🤔

Artificial general intelligence is not inevitable. My latest for The Guardian challenges one of the most popular claims made about AGI. Among those who believe AGI is possible, it's common to think it's unstoppable, whether you're excited or terrified of the prospect 🧵

Reposting 19 month old screenshots w/o context is just misinforming people at large scale. (The free version of ChatGPT gets this right btw.) Plenty of good reasons to not like AI, but tricking people into thinking it's dumber than it actually is bad. x.com/racistmarx/...

Anthropic could be bankrupted within the next few months, thanks to last week's barely covered legal ruling, which exposes the AI startup to billions to hundreds of billions in damages for its use of pirated, copyright-protected works.

New reporting from WSJ. Federal contract eligibility has been an underrated backdoor into regulating AI, given how big a customer the feds are. The parallel with CCP AI regs — which focus a lot on political correctness of a different sort — is interesting.