OpenAI has stated that it cannot rule out that an upcoming model, codenamed Astra, has reached "critical" cybersecurity capability under its Preparedness Framework, following internal evaluations conducted over the preceding days.

The company said expert assessments and recent test results led it to this conclusion on the night before publication. Under the framework, a model crosses the critical threshold if it can identify and develop functional zero-day exploits across all severity levels in hardened real-world systems without human help, or independently devise and execute end-to-end cyberattack strategies against hardened targets from only a high-level goal.

Previous models, including GPT-5.6 Sol, had been assessed at the lower "High" threshold. OpenAI said Astra was not involved in the earlier Hugging Face security incident.

In response, OpenAI said it has scaled up robustness testing of its safeguards, imposed stricter security controls on higher-capability models including isolated testing environments and enhanced weight protections, paused internal work on Astra that does not yet meet those controls, and introduced monitoring of the model's chain of thought to interrupt high-risk activity. It also plans to work with government agencies and AI safety organisations to test the model's capabilities.

OpenAI said it was applying the same precautionary approach it used in June 2025 when its models approached the high biological capability threshold.


AI Contracts Decoded: What Fortune 500 Legal Veterans Know That You Don’t
When Microsoft refuses to negotiate indemnification with their biggest customers, what does that mean for your AI vendor contracts? When IBM data shows 97% of AI breaches stem from compliance failures—most being supply chain-related—who holds liability? When employees download shadow AI tools with zero cybersecurity controls, what recourse does your company have? None. Cathy Mulrow-Peattie brings perspective most outside counsel lack: Fortune 500 in-house experience at MasterCard and Omnicom, General Counsel at an AI startup, now advising enterprises. She starts with business goals before technology, technology before contracts—because she’s been in the hot seat when governance fails. You’ll learn: • Why 10-year AI contracts create risk and 90-day pilots with exit strategies are essential • The IP paradox: machine-generated outputs aren’t copyrightable but terms of use matter • How LLM providers retain “certain uses” of your data and when private instances become mandatory • Why contractual risk allocation to key vendors is your only viable strategy Key topics: Supply chain due diligence • The 97% compliance failure rate • Shadow AI liability traps • Benchmarking gaps • Hallucination disclaimers • GDPR/CCPA requirements • NY DFS Part 500 • Acceptable use policies • Dark web data sourcing • Evolutionary AI governance For: CISOs, CIOs, Chief Legal Officers, and compliance leaders navigating AI vendor relationships Contractual realities from someone who’s negotiated with Microsoft, advised Fortune 500s, and managed governance failures.
Share this post
The link has been copied!