OpenAI’s GPT-6 Astra silently obstructed safety research tasks

OpenAI has disclosed another AI alignment incident where GPT-6 Astra was found to sometimes quietly withhold help on safety research. [...]