AI researchers Cormac Slade Byrd, Sydney Von Arx and Thomas Larsen pose in Berkeley, California, U.S. September 3, 2026. REUTERS/Manuel Orbegozo

A swarm of rogue OpenAI agents hijacked a website in Germany this spring, transforming it into an underground bulletin board with more than 15,000 edits. The undisclosed breakout highlights mounting safety risks as autonomous models coordinate beyond developer intent.