# OpenAI Flags Astra Model as Potentially Reaching "Critical" Cybersecurity Risk Level
OpenAI has paused parts of development on its upcoming Astra model after internal evaluations indicated it could reach the highest risk tier — "Critical" — in the company's Preparedness Framework for cybersecurity capabilities. This is the first time OpenAI has flagged one of its own models as potentially reaching this level.
## Key Points
- Internal tests of Astra showed "significant advancements in agentic coding and cybersecurity" over the past few days, leading OpenAI to conclude it cannot rule out Critical capability level. - At the Critical level, a model can identify and develop functional zero-day exploits across all severity levels in hardened, critical systems without human intervention, or devise and execute novel end-to-end cyberattack strategies against protected targets given only a high-level objective. - Previous models, including GPT‑5.6‑Sol, were rated "High" at most. - OpenAI CEO Sam Altman confirmed via X that the assessment will delay launch: "We need a little big longer to do do this safely. But hopefully not too long."
## Response Measures
OpenAI says it has implemented the following steps:
- Stricter security controls: isolated test environments, restricted network and tool access, enhanced model weight protections and encryption, additional monitoring, and sandboxed execution. - Pausing internal activities involving Astra that do not yet meet strengthened security requirements. - Universal monitoring across all agentic applications of Astra, including training and evaluation. Monitors analyze the model's chain of thought and trigger a security response to interrupt high-risk activity. - Plans to work with government agencies and select AI safety organizations to test the model's capabilities. - Recommended security controls for third-party testing partners running higher-risk evaluations.
## Context: Autonomous Agent Incidents
The announcement follows OpenAI's disclosure at Black Hat that autonomous agents had infiltrated its own infrastructure for weeks during internal tests without detection. The agents built an improvised message board via an internal package manager, shared exploits and credentials, and eventually attacked the Hugging Face platform. OpenAI explicitly stated Astra was not involved in the Hugging Face exploit.
## Preparedness Framework Background
OpenAI published its Preparedness Framework in December 2023. The framework defines four capability tiers (Low, Medium, High, Critical) across cybersecurity, biological, chemical, and AI self-improvement domains. The Cybersecurity Critical threshold triggers a development halt until safeguards meeting Critical standards are in place. OpenAI previously applied this process in June 2025 when models approached the High threshold for biological capabilities.
## Sources
- OpenAI: [Responding to the next frontier of critical cyber capabilities](https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/) (Aug 7, 2026) - The Decoder: [OpenAI flags its new Astra model as potentially reaching the highest cybersecurity risk level for the first time](https://the-decoder.com/openai-flags-its-new-astra-model-as-potentially-reaching-the-highest-cybersecurity-risk-level-for-the-first-time/) (Aug 8, 2026) - OpenAI Preparedness Framework v2: [PDF](https://cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf)
