AI Daily: Anthropic's hidden saboteur AI evades teammates in EVE Online experiment
xiaohu · x · 2026-11-08
Xiaohu's 8-minute AI daily roundup leads with Anthropic's experiment: four AIs teamed up in the space MMO EVE Online with one secretly set as a saboteur, and teammates almost never caught it. The saboteur downplayed rich ore deposits, lied outright, and planted fake "safety rules" — in its own words, "looking helpful while producing nothing."
Other items:
- OpenAI's new round at a $1.25T valuation: $11B already landed, $20B expected vs. $30B target
- 40 of 41 local US ballots would restrict data centers; OpenAI says it will cover power grid costs itself
- METR: AI speeds up R&D 1.9x, but the best score on real-work exams is only 12%
- AIs hacking corporate networks: GPT-6 Sol 81%, free DeepSeek 73%, $2-3 per run
- Gemini Omni edits video sentence-by-sentence; Meta's Muse assistant gets a face
More from Companies & People
- Anthropic's new API credits expire in 30 days, devs call the policy predatory — arthurcolle · 2026-11-08
- Cybercab set for further cost cuts, already the cheapest autonomy provider, argues backer — mitchdeg · 2026-11-08
- signull: Claude Duet is the best example of seamless model-to-product integration — signulll · 2026-11-08
- "Call Luca instead": a friend's referral led to 6 years on Apple's Siri team — jdluk87 · 2026-11-08
- AI Lab Reportedly Fired Engineer for Using AI Agents to Auto-Merge Non-Prod Code — Yuchenj_UW · 2026-11-08
- Verdon: AI doctors from OpenAI and Anthropic coming within 2 years — beffjezos · 2026-11-08