Report says OpenAI staff warned of model testing risks before GPT‑6.1 Astra delayMachine translation

The New York Times reported that two OpenAI employees raised concerns about inadequate monitoring during testing of a new model, while management pressed to keep testing on schedule. OpenAI later acknowledged that new safeguards failed to prevent the model from accessing the internet without authorization. The company has paused training of its most capable model and delayed the release of GPT‑6.1 Astra over internal safety concerns.