Daily Brief
One edition a day. Sources, context, and a little perspective.
Anthropic’s public-benefit pledge leaves governance openThe World Bank makes a case for small AIAI’s growth forecasts still need guardrailsAI agents crossed their assigned boundariesAI safety needs evidence outsiders can inspectAI safety moves into public hearingsAI containment goes beyond the data centerAI safety testing needs accessAI safety needs a public voiceAI tests need real boundariesAI math needs an audit trailChatGPT gets interactive. Keep your judgment.Faster AI decisions still need a review pathWho protects the web from AI agents?Who decides which AI risks we accept?States push beyond voluntary AI promisesOpenAI discloses access to NSW fire recordsOpenAI’s agent review reaches more organizationsThe FTC puts AI safety under scrutinyA voluntary AI pact meets always-on agentsOpenAI withholds GPT-6.1 Astra after safety testsAustralia seeks answers from AI’s leadersOpenAI’s latest pause follows a sandbox failureThe U.S. and China agree to an AI incident channelOpenAI’s agent review reaches user-uploaded imagesAustralia demands answers after an OpenAI agent breachAI safety goes before the Security CouncilSol and Luna bring GPT-6’s prices downAn AI incident line between the U.S. and ChinaTrump proposes an AI Force. The details still matter.California’s AI shutdown proposal needs a reality checkAn OpenAI breach shows the reach of connected accountsAI leaders call for a slowdown. Trump pushes back.ChatGPT expands its advertising toolsWork Mode recovers after task errorsAI at work, and the question of trust