Three safety researchers were dismissed for allegedly sharing confidential company information with an outside AI safety organization, OpenAI confirmed — the same week the company shelved its new GPT-6.1 Astra model over safety concerns.