1,178 Employees at OpenAI, Anthropic, Google DeepMind and Meta Sign Statement Urging US to Help Pace AI Development
- 1,178 employees from frontier AI labs including OpenAI, Anthropic, Google DeepMind, and Meta signed a public statement asking the US government to back an international effort to build tools for deliberately pacing automated AI development.
- Signatories include high-profile figures such as Dario Amodei (Anthropic CEO), Jakub Pachocki (OpenAI Chief Scientist), Jared Kaplan (Anthropic co-founder), Jack Clark, Chris Olah, and Jan Leike.
- The statement argues labs believe they are close to automating AI research itself, which could accelerate capability gains beyond current ability to understand or control the resulting systems.
- Signers like Leo Gao and Dawn Song cite concrete evidence, including CyberGym and ExploitGym evaluations showing frontier AI agents can already discover and exploit real software vulnerabilities.
- The statement stops short of naming specific mechanisms, only calling for governments to develop unspecified technical and governance tools, which HN commenters criticized as vague and possibly self-serving regulatory capture.
Hacker News 의견들
Compute access doesn't give these guys special insight into the future of AI. Literally everyone with enough GPUs is shipping models now, why should they be gatekeepers.
This statement is way too vague, what does 'pace the frontier' actually mean in practice?
It means only vetted actors get to use frontier models, basically don't share the frontier broadly.
I read it as wanting real infrastructure: international agreements, GPU-level verifiability, physical datacenter inspections, like what we do with nuclear material.
Could be regulatory capture if you're cynical, but I think a lot of these people genuinely believe it. The real fear isn't Skynet, it's bots and bad actors misusing compute for scams and dark patterns, that's 100% a human problem not an AI-comes-alive problem.
The site only shows about 50 of the 1100+ comments and won't let you load the rest. Also weird how many em dashes show up across supposedly different signatories' comments.
Pacing already happens through economics. Cost-per-token decides what tasks actually get automated, a gas station clerk's job is safe because the tokens to replace them cost more than their wage.
Sure, token economics might pace adoption into the labor market, but that has way less influence on how fast the actual frontier models improve.
Pacing is a bad idea because China won't slow down to match you.
Exactly, do these companies show any real concern beyond wanting to slow down now because they're losing the race?
The whole point of the statement is literally to kickstart international coordination so nobody has to unilaterally fall behind.
Every early computer was built for war, the Mark I did implosion physics for the bomb, ENIAC ran H-bomb calculations. If people in 1955 had campaigned to restrict computing because it 'only makes weapons,' we'd have never gotten here.
Same pattern happened with the printing press and the internet, people in power always panic that the masses will misuse new tools. Bad stuff did happen, but trying to bag either invention would've been a fool's errand given the long-run gains.
This is just standard AI doomer messaging dressed up as a compliment sandwich, 'AI is so powerful that's why we must slow down.' The actual boring risk nobody mentions: AI is about to gut software dev jobs, including hundreds of thousands of visa workers, and these companies say nothing about that ethics problem.
If you're serious about not contributing to 'problematic progress,' just quit your job. Curious why none of these signatories do that.
That's a tragedy of the commons though, one person quitting doesn't stop the frontier, someone else just fills the seat. That's exactly why you need a binding rule that forces everyone to slow down together.