What changed: OpenAI delayed the release of GPT-6.1 Astra, according to reporting by The Associated Press. The company said its researchers had raised security concerns as the model became more persistent at completing tasks.
AP reported that OpenAI safety systems head Saachi Jain said the version did not meet the company's bar and that additional safeguards were needed before release. The report also said OpenAI had paused training of its most advanced models while it evaluated protections.
Why it matters: This is a concrete example of a frontier lab treating agent persistence as a deployment question, not simply a benchmark improvement. For organizations planning around model roadmaps, the immediate lesson is that safety testing can change release timing—especially for systems that can act over longer task chains.