OpenAI, Anthropic Rethink Independent Safety Evaluations

Weeks after agreeing to a voluntary program allowing outside evaluators into their labs, OpenAI and Anthropic are seeking to limit what evaluators can disclose, fearing proprietary information could reach rivals. Measures under consideration include monitoring embedded evaluators, barring them from working at multiple labs, and reconsidering editorial control over publications.

• Five ways to make the new US AI task force succeed2026-10-07
• Commentary: Preventing AI from triggering a world war2026-10-05
• This week in AI: Trump, Zuckerberg, Musk and Nvidia2026-10-04
• Trump pledges an 'AI force' and a dedicated AI czar2026-10-02
• Trump's 'super intelligence' accord includes spelling error2026-10-01