OpenAI has begun a phased rollout of its latest AI model, GPT-6 Astra, starting with some companies in its application-based cybersecurity programme, Daybreak. Astra is the first model to reach a \'critical\' level under OpenAI’s internal cybersecurity standards, and access to its advanced security functions will be restricted. The rollout comes after recent security incidents that prompted OpenAI to strengthen safety controls and apply additional safeguards to Astra.

OpenAI has unveiled a new safety system aimed at monitoring AI misuse without storing enterprise customer data. The company is previewing Private Safety Processing to some customers, testing a method that links multiple chat sessions to detect risky signals while keeping data under its Zero Data Retention policy. OpenAI said agents can analyze patterns across sessions without direct human review of user conversations. The move contrasts with Anthropic’s policy allowing 30-day retention for certain models.