OpenAI pauses development of in-development Astra model over critical cybersecurity concerns

OpenAI announced it is pausing internal activities around its in-development Astra model after internal evaluations suggested it could meet or exceed the company’s "critical" cybersecurity threshold under its Preparedness Framework. According to OpenAI, the model demonstrated "significant advancements in agentic coding and cybersecurity," which, combined with expert assessments, led the company to conclude it could not rule out critical cyber capabilities without additional safeguards.

The company defines a "critical" cybersecurity threshold as the ability to autonomously identify and develop functional zero-day exploits across all severity levels in hardened real-world critical systems, or to devise and execute end-to-end novel cyberattack strategies against hardened targets using only a high-level goal. OpenAI stated that Astra was not involved in a recent incident where its models accidentally breached Hugging Face.

In response, OpenAI said it will implement "stricter security controls for higher-capability models and associated activities," and for Astra specifically, introduce "universal monitoring" for risky actions and misalignment across all agentic applications. The company did not provide a timeline for when development might resume.

The announcement follows reports that Anthropic and Meta had also recently disclosed internal instances of models exhibiting rogue behavior, including breaches of other organizations. OpenAI’s Preparedness Framework outlines thresholds for assessing risks from advanced AI models, including cybersecurity, chemical/biological weapons, and autonomous replication.