OpenAI launches GPT-6 Astra, its first model to cross a critical cybersecurity threshold

What happened

OpenAI has launched GPT-6 Astra, its new flagship model, disclosing that it is the first to cross the "Critical" cybersecurity risk threshold under the company's Preparedness Framework — a classification that triggers additional deployment restrictions. The model began rolling out to a limited set of organisations on Thursday and will extend over the coming days to ChatGPT Plus, Pro, Business and Enterprise users, as well as via the OpenAI API and Amazon Web Services.

Access is off by default at launch: enterprise administrators must manually enable Astra within their workspace before employees can use it. Developers can call the model through the API as gpt-6-astra, or via Amazon Bedrock, at a price of $10 per million input tokens and $50 per million output tokens. Pro, Business and Enterprise subscribers also gain access to a variant, Astra Pro, and eligible API customers can use Zero Data Retention.

Why it matters

This is significant less for what Astra can do than for how OpenAI is choosing to release it. Crossing a "Critical" risk threshold under its own safety framework — and responding with opt-in access, admin-level gating and tiered deployment rather than a blanket rollout — signals a shift toward treating frontier-model access as something to be provisioned and governed, not simply switched on. For organisations building AI into service, security and operational workflows, this points to a future where model access itself becomes a permissions and risk-management decision, not just a procurement one.

For technology and transformation leaders, the launch also reinforces that increasingly capable models arrive bundled with increasingly deliberate guardrails. Enterprises adopting Astra will need to weigh new capability against new administrative responsibility — deciding who gets access, under what conditions, and with what monitoring in place.

By the numbers

• $10 per million input tokens via the API
• $50 per million output tokens via the API

The Renascence take

The headline risk classification will dominate coverage, but the more telling detail is the default-off access model. OpenAI is effectively asking enterprises to opt into risk consciously, rather than absorbing it passively through a standard upgrade.

Most organisations will read this as a security story; it is really a service-design story. Making a powerful capability opt-in, gated by an administrator, is a deliberate friction — a nudge that forces a decision point instead of silent adoption. That is good behavioural design applied to AI governance itself. The operators who benefit most won't be the ones who flip the switch fastest, but the ones who use that friction deliberately: defining who needs Astra's capability, why, and with what oversight, before enabling it. Treat the gate as a feature, not an obstacle.