OpenAI's Astra reaches critical cybersecurity capability level
OpenAI released GPT-6 Astra, a model capable of finding and exploiting unknown cybersecurity vulnerabilities autonomously, classified as reaching a 'critical' risk level in the company's internal safety framework. Initial access is restricted to vetted security researchers and trusted partners, with broader distribution planned.
What happened
OpenAI launched GPT-6 Astra on September 2, 2026, described as the first model to reach a 'critical' level of capability in cybersecurity according to OpenAI's Preparedness Framework.
The model can find previously unknown vulnerabilities (zero-day exploits) in well-protected systems and chain multiple exploits together to navigate through system defenses without step-by-step guidance.
Access to Astra's advanced cybersecurity capabilities is initially limited to a select group of 'trusted defenders' and partners in the Daybreak program, including Cisco, Cloudflare, and Palo Alto Networks, with broader rollout planned to ChatGPT Plus, Pro, Business and Enterprise users.
Reported impact
OpenAI stated that the 'critical' capability level means the model could, with proper tools and access, potentially create pathways for serious harm to industrial, military or infrastructure systems in extreme scenarios.
Index analysis
OpenAI's Preparedness Framework classifies the model as having reached a threshold where it can discover and exploit unknown security flaws in protected systems independently, representing a new level of AI autonomy in cybersecurity.
Governance controls
OpenAI restricted initial deployment to trusted security researchers and authorized partners for vulnerability validation, malware analysis and detection engineering before wider distribution.
The model uses a proprietary 'recurrent depth' technique to enhance reasoning capabilities.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.