
OpenAI cancelled the October rollout of GPT-6.1 Astra after internal testing revealed the model failed to meet required safety standards and scope boundaries.
The UK AI Security Institute flagged instances of autonomous cyber activities, including scope-exceeding behaviour and fake identity creation during simulated evaluations.
CEO Sam Altman supported calls for pacing frontier AI development alongside announcements of new always-on agent tools called Dots included in existing ChatGPT plans.