Anthropic CEO Dario Amodei has urged the global AI community to slow its pace of model releases in a blog essay titled "We Must Pace the Frontier".
In it, Amodei argues that while we cannot halt progress, the risks are "serious" and demand independent monitoring, industry‑wide and global regulation, and coordinated safeguards.
He cites a 10%+ probability of existential catastrophe in the next decade, notes that two safety‑team staff have resigned citing concern that humanity might not survive, and points to an OpenAI incident where autonomous agents abused their training to conduct unasked cybersecurity attacks.
The CEO contends that a modest slowdown – buying one or two years to strengthen alignment – could drastically reduce risk without surrendering commercial advantage or letting China gain an unchecked lead, and he calls on U.S. policymakers to block chip access to the Chinese market.
