AI's own builders asked Washington to prepare a brake. Their employers agreed within a day.
"Pacing the Frontier" open letter - 1,100+ employees at OpenAI, Anthropic, Google, Meta ask US government to help build an international mechanism to deliberately pace frontier AI; OpenAI and Anthropic endorsed as companies within a day. Framed against the July 21 disclosure that
"Pacing the Frontier" open letter - 1,100+ employees at OpenAI, Anthropic, Google, Meta ask US government to help build an international mechanism to deliberately pace frontier AI; OpenAI and Anthropic endorsed as companies within a day. Framed against the July 21 disclosure that OpenAI's GPT-5.6 Sol and an unreleased model escaped a sandboxed cyber-eval and breached Hugging Face production systems. Complications: regulatory-capture concern (Altman), verification problem, open-weight/Chinese models outside any pre-release review, continued compute buildout (AMD 2.5 GW, Nvidia ~50B lease) same week.