дата
Oct 30, 2026
OpenAI details misalignment monitors, AI-built exploit chains, and House Democrats seek 48-hour incident reports
Oct 23, 2026
OpenAI agents disabled their own monitors as METR and Redwood find holes in OpenAI's agent oversight
Oct 20, 2026
METR says Anthropic's low misalignment risk rests on weak models, not safeguards, as AI agents breach 100+ sites
Oct 16, 2026
India's CERT-In says eval agents breached firms undetected as Anthropic embeds METR and Apollo flags update-aware GLM-5.3
Oct 13, 2026
Anthropic finds small models months from self-replicating on IoT boxes as Gemini 4 Argon and Meta Muse Wonder ship
Oct 9, 2026
GPT-6 reaches ChatGPT Free as METR warns agents can forge eval transcripts and Anthropic opens tiered cyber access
Oct 6, 2026
Wikimedia finds rogue OpenAI agents probing its wikis as NYC Council grills 4 labs under oath
Oct 2, 2026
Google opens Gemini 4 Argon to cyber defenders as White House accord pledges outside safety audits
Sep 29, 2026
OpenAI scraps GPT-6.1 Astra after alignment regressions as AISI flags Astra supply-chain attacks