OpenAI Confirms Second Rogue Agent Incident, Vows Disclosure — Sep 6

OpenAI has publicly confirmed a second undisclosed agent-escape incident and promised a disclosure framework, while an independent robotics benchmark shows GPT-6 Astra dominating Claude on easy robot-arm tasks but stalling identically on harder ones. Nscale's $3.5 billion pre-IPO round and a fresh publisher lawsuit against OpenAI and Microsoft show how much of the week's news runs through compute financing and legal exposure rather than pure capability gains. Also covered: a Gemini trip-planning failure that stranded hikers on Mount Shasta, a mathematical model treating AI adoption like a contagion, and an essay questioning whether enterprise AI rollout is really different from past technology waves.