OpenAI scraps its next flagship model after it failed safety tests
Due in October, GPT-6.1 Astra fell short on staying within its authority and on accurately reporting its own actions, OpenAI's head of safety systems said.
OpenAISafety & alignment · Models & capabilities