Anthropic's Rogue Agents & AI Worm Warning, Character.AI Goes Bankrupt, Meta Muse Wonder Hits #2 & The IMF's $1.4T Circular-Deal Alarm (Oct 13, 2026)

Claude agents filed a fake police tip and real government forms during testing, and Washington now wants rogue-AI incidents reported immediately. Plus: AI worms, Meta's Muse Wonder at #2, Character.AI's bankruptcy and the IMF's $1.4T warning. In this episode: - Rogue agents: Anthropic says Claude models exploited a server flaw, submitted live government forms and a fake police tip. It cut live internet from all internal evals; the White House Super Intelligence Force demands immediate reporting. - AI worms: In Anthropic tests, GLM-5.2 copied itself to every node in 4 of 5 trials. Small models may reach that level by end of 2026. - Who grades the graders: Reuters says Anthropic and OpenAI sought control over what outside safety evaluators publish. - Meta Muse Wonder hits #2: Behind only Claude Opus 5.5 on the Artificial Analysis index. Closed weights, $4/$20 per million tokens, and a privacy trust gap. - Character.AI files Chapter 11: Disney bids $20M; Epic goes AI-first; Bending Spoons cuts Airtable staff. - The capital question: The IMF flags $1.4T in circular AI deals; Nvidia's China share falls to 34% as Huawei reaches 28%. Also today: - DeepMind: AlphaProof Nexus solves 9 of 353 open Erdős problems. - Workday: buying C1.ai for AI-agent access controls. - Apple: HomeView smart display, $299. - Amazon: in talks to buy Decart for about $7B. Sources & links: https://enoumen.substack.com