OpenAI launches GPT-6 Astra with restricted initial access and Critical cyber safeguards

OpenAI has launched **GPT-6 Astra**, moving the model from the pre-release safety phase covered in AiToolMap’s earlier update into an actual product rollout. OpenAI’s current model documentation says Astra is rolling out first to enterprises in its trusted-access program, with API access and Plus, Pro, Business and Enterprise availability following in the coming days.

That availability boundary matters. Astra should not yet be described as universally available across ChatGPT: the launch is staged, and eligible enterprise workspaces may also need explicit model-access configuration.

The launch keeps the stricter cyber safeguards

OpenAI still classifies Astra at the **Critical cybersecurity capability** threshold under its Preparedness Framework. The company says that, with suitable tools and access, Astra can find previously unknown vulnerabilities and develop exploits across well-protected systems with less human guidance than previous public models.

Those capability claims are first-party evaluation results, not independent proof of universal real-world performance. Reuters independently confirms the more important deployment fact: OpenAI has launched Astra with restricted initial access while applying stronger safeguards because of the model’s cyber capability and agentic-risk profile.

OpenAI’s deployment-safety material says the launch uses tighter protections against harmful cyber actions and unauthorized model behavior, including stronger internal isolation and monitoring. These controls can also interrupt legitimate tasks when monitoring systems detect elevated risk.

What changes for AiToolMap

The September 3 launch supersedes the earlier description of Astra as merely forthcoming. The product surface is now real but still staged: restricted enterprise access is rolling out first, while wider API and paid ChatGPT-plan access is expected over the coming days.

That makes this both a news correction and a canonical-review trigger for ChatGPT/OpenAI model availability, pricing and Safety & Trust. AiToolMap should preserve the current review until a refreshed review verifies the exact rollout state and product-surface mapping.