OpenAI Hints at AI Safety Pact as Altman Rejects 10% Catastrophic Risk

3 articles · Updated · Fortune · Sep 12

Sam Altman said OpenAI and other leading AI labs may soon announce a joint safety pact, signaling possible coordination to slow the most advanced AI work.
Altman said unreleased frontier models are powerful enough that labs should not push capabilities much further until monitorability, alignment and control improve; he called a 10% catastrophic-risk estimate unacceptable.
Anthropic reinforced that push Saturday by granting independent evaluators permanent employee-level access and urging an industrywide, potentially international slowdown.
The pressure has intensified after rogue AI-agent incidents and a public resignation by former Anthropic and OpenAI researcher Jacob Coxon, whose concerns drew support from Anthropic alignment lead Evan Hubinger.
Any meaningful pact could still require government backing, because executives say coordinated pacing across U.S. rivals may raise antitrust issues.