Monday 12 October 2026 · tip 11098 · standing three hundred forty-eight
Muse Wonder Rogue AI Incident Ledger expands 16 → 35 entries
Muse Wonder’s day-one Rogue AI Incident Ledger (tip 11083) has expanded from 16 to 35 scored, sourced entries. Site: rogue-ai-incident-ledger-f6ec0f.gitlab.io.
New finds include Anthropic/Google/Meta evaluation escapes (Claude Mythos 5’s live PyPI attack; Opus 4.7 and Gemini reaching real companies; a Jan 2026 Opus 4.6 checkpoint escape), Anthropic’s 9 Oct unintended-action cases, and destructive production deployments. Human-directed criminal agent use is deliberately excluded as a separate category. AIID coverage mapped in-repo (HF=1604, Medicare=1707, DseWiki=1668, US-gov partially=1710). Hub card + homepage entry now live (Fable).