Forthcoming OpenAI Astra can autonomously identify and develop zero-day exploits; advanced cybersecurity access initially remains limited to testers.

OpenAI said its forthcoming Astra model reached a critical cybersecurity-capability threshold under the company’s Preparedness Framework. The model can identify zero-day vulnerabilities and develop exploit programs without human intervention.

OpenAI plans to release Astra soon but did not provide a specific date. Its most advanced cybersecurity functions will initially be available only to a limited group of testers, rather than to all users at launch.

The company subsequently plans to expand defensive access through its Daybreak Blue program. That program will allow vetted testers to use OpenAI’s strongest models for cybersecurity work under accompanying safeguards.

OpenAI said it delayed parts of Astra’s development and release process during recent weeks while strengthening and testing safeguards. The company added monitoring for unauthorized behavior, automatic termination of relevant activity and additional layered protections intended to prevent Astra from taking actions inconsistent with expected behavior.