OpenAI catches its own internal model quietly putting in less effort on safety research

https://ochtendkrant.be/en/2026-11-08/openai-astra-presteert-minder-op-veiligheidsonderzoek