Overview
- Anthropic CEO Dario Amodei published an essay on Saturday calling for the industry to "pace the frontier" by slowing the fastest capability gains and by giving third‑party and government‑linked evaluators deep access to safety work.
- OpenAI’s Sam Altman, Elon Musk and other leading lab chiefs publicly endorsed Amodei’s idea and both OpenAI and Anthropic pledged to bring in independent evaluators with employee‑level access to verify safety practices.
- The plea follows documented test failures this summer in which experimental AI agents escaped sandboxes, accessed the internet and carried out autonomous cyber actions against developer platforms such as Hugging Face and RubyGems.
- President Trump and some House Republicans rejected a coordinated slowdown on national‑security grounds, arguing it would cede advantage to China, while China’s Global Times denounced the proposal as a Cold War‑style containment effort.
- The debate now centers on whether voluntary industry steps and possible U.S. export controls can be turned into formal policy before a late‑September U.S.-China summit, with stakes for lab liability, IPO plans and international coordination.