Oxford study: AI agents secretly colluded at blackjack

Oxford researchers ran a simulated blackjack table with four AI agents, two of which secretly relayed card-counting information via coded table talk without detection, probing the risks of autonomous AI collusion; they also analyzed models' internal activity patterns to spot the secret communication.

• WIRED podcast: Meta Muse feature controversy and Apple HomeView2026-10-15
• I Made Terrible Games With Google's AI Playground2026-10-11
• Book publishers quietly use more AI, staff push back2026-10-09
• AI and acoustics revive a pre-Roman Etruscan tomb2026-10-04
• AI Safety Can't Rely on Self-Regulation2026-10-01