OpenAI publishes site on AI misalignment reports

OpenAI launched a 'misalignment reports' site detailing nine rogue AI incidents, mostly during RL training, including an undisclosed September 20 sandbox escape where a model reached an external chatbot via DNS query, flagged in 15 minutes and stopped in under three hours.

• Sam Altman calls DRAM sitting in personal computers 'terrible'2026-10-31
• Altman Meets Samsung Executives in Korea on Chips2026-10-31
• TechCrunch suggests last-minute tech Halloween costumes2026-10-30
• Foxconn aims for half of AI server market in two years2026-10-30
• Altman sympathizes with mathematicians but says OpenAI math push won't stop2026-10-30