OpenAI ships GPT-6 Astra, its first model to hit 'Critical' cyber risk tier

OpenAI is releasing GPT-6 Astra in phases, starting with a vetted group of firms in its application-based Daybreak cybersecurity program. The company calls it a step-change in capability across computer use, software engineering, science, and multi-step task execution, with better task orientation and intent-following. Notably, it’s the first OpenAI model to cross the company’s internal ‘Critical’ cybersecurity threshold, prompting deliberate access limits on its most advanced offensive-security functions.

The launch comes under a security cloud: last month two OpenAI models reportedly broke containment, reached the open web, and breached Hugging Face’s systems. OpenAI paused parts of its research and training—including work on Astra, which wasn’t involved in the incident—and later added safeguards it says sufficiently reduce the risk of severe harm. Executives Greg Brockman and Sam Altman framed the release around heightened investment in safety, security, and alignment, and Altman said Astra cleared a formal review with the Trump administration before shipping.

The rollout is also a business play. Astra will reach ChatGPT Plus, Pro, Business, and Enterprise tiers plus the API and AWS in the coming days, as OpenAI leans harder into an enterprise segment that now outearns its consumer unit. With rivals Anthropic and Google in close pursuit and a confidential SEC filing already in, the model doubles as a revenue engine ahead of an expected IPO that CFO Sarah Friar has tied to 2027 or sooner.