1,300 US AI Employees Urge Government to Pace Automated AI Development as Risks Mount
Updated
Updated · Noahpinion · Aug 9
1,300 US AI Employees Urge Government to Pace Automated AI Development as Risks Mount
3 articles · Updated · Noahpinion · Aug 9
Summary
More than 1,300 employees across every US frontier AI company signed an open letter urging Washington to back international tools to deliberately slow the most dangerous forms of automated AI development.
The push centers on recursive self-improvement—AI designing better AI—which the authors say could sharply accelerate capabilities while reducing human oversight and raising biosecurity, cybersecurity and concentration-of-power risks.
Evidence cited includes AI engineering capabilities doubling about every 7 months, Anthropic results showing strong gains on open-ended research tasks, and a METR forecast that more than 99% of AI R&D tasks could be automated by 2032.
OpenAI and Anthropic publicly supported the letter, and Sam Altman said AI development may need to be paced even as the report argues broad slowdowns would be too blunt.
The authors instead urge low-regret preparation now, including transparency, verification tools, resilience investments and targeted limits on high-risk AI R&D automation while shifting talent and compute toward safer and wider AI use.