OpenAI starts GPT-6 Astra rollout with critical cyber safeguards

OpenAI has started rolling out GPT-6 Astra, its newest frontier model, to a limited set of organizations before wider availability for ChatGPT Plus, Pro, Business and Enterprise users, the API, Microsoft Azure and AWS Bedrock.

The company is positioning Astra as a broad jump in computer use, coding, browsing, science and professional work. OpenAI says the model completes OSWorld-style computer-use tasks faster than GPT-5.6 Sol while scoring higher, and that the updated Codex harness plus Astra improves task-completion speed on Mind2Web by about 1.9x.

The important caveat is cybersecurity. OpenAI says Astra meets the Critical threshold in cybersecurity under its Preparedness Framework. In testing without production safeguards, the model reached 100% on ExploitBench, improved on ExploitGym, and found two previously unknown vulnerabilities that OpenAI says it is disclosing to maintainers. The launched version keeps refusals for advanced offensive tasks such as creating proof-of-concept exploits.

OpenAI is pairing the launch with Daybreak for Frontline Defenders, a $1B commitment to subsidize frontier cyber AI access, training and support for organizations that protect essential services. The program starts with U.S. water, electricity, local government, banking, nonprofit and open-source defenders, and includes a pilot with MS-ISAC plus more than 35 partner products and services.

For users, Astra is a major model launch. For security teams, OpenAI is also signaling that frontier cyber capability is moving from warning label to controlled deployment.

Update: OpenAI has temporarily paused new sign-ups and upgrades to the $200 ChatGPT Pro plan after Astra demand put pressure on system capacity. Existing Pro accounts are unaffected, and OpenAI says the lower $100 Pro tier, API, Business and Enterprise access remain open while it adds capacity.