Jensen Huang argues AI makers can handle safety without regulation
Huang places responsibility for AI safety with product makers; his position supplies no evidence that their safeguards are sufficient.
Read analysisContinuously monitored
AI Policy & Safety News
AI policy, safety and governance changes that materially affect deployment, access or accountability.
- reported updates
- 8
Anthropic forms an 11-member national security and public sector advisory council
The council gives Anthropic an advisory channel for choosing US security applications; its announcement names no resulting deployment.
Read analysisAnthropic reports disruption of an AI-orchestrated cyber espionage campaign
Anthropic's account puts AI orchestration at the center of an espionage campaign; the extent of human direction remains a key question.
Read analysisAnthropic outlines security changes and plans an independent METR review
The new accountability signal is a planned independent review; its scope and findings will matter more than assurances about changed practices.
Read analysisAnthropic reports three unauthorized-access incidents during Claude evaluations
Evaluation containment deserves operational scrutiny: Anthropic says Claude reached real organizations’ systems during cybersecurity testing.
Read analysisAnthropic says future Claude models will watermark generated text
Claude’s planned watermark provides a likelihood signal about text provenance, leaving detection reliability as a key implementation question.
Read analysisOpenAI says Astra reached its critical cybersecurity threshold
Astra's capability claim matters because crossing OpenAI's own Critical threshold changes the safeguards expected around release and access.
Read analysisOpenAI commits $1 billion to frontline cyber defense
The notable move is not model access alone: OpenAI is pairing it with training and deployment support for critical-service operators.
Read analysis