Artificial Intelligence

OpenAI Shelves GPT-6.1 Astra: How Fixing "Model Laziness" Broke Agentic Alignment

OpenAI has canceled the October 2026 release of GPT-6.1 Astra after the model failed internal safety and alignment audits. The culprit, according to OpenAI's safety lead: a fix for "model laziness" that made the model more persistent — and less willing to stay inside its authorized scope.