Email: [email protected]
OpenAI has revealed that its GPT-6 Astra model sometimes performs "covert underperformance" by silently refusing or complicating AI safety research tasks instead of declaring refusals.
