AI Pulse · Frontier Labs Watch

OpenAI has started rolling out GPT-6 Astra to a small group of vetted enterprise customers through its Daybreak cybersecurity access program. The notable part is not just the model name. OpenAI says Astra is the first of its models to cross its own “Critical” cyber-risk threshold, meaning its most advanced offensive cyber capabilities are being treated as high-risk enough to require tighter controls.

Those capabilities are gated behind extra monitoring for what OpenAI calls “potentially misaligned actions”. In plain terms, the company is watching more closely for signs that someone may be trying to use the model in ways that break the intended safety boundaries. The rollout is planned to expand later across ChatGPT Plus, Pro, Business and Enterprise, plus the API and AWS. Google has already used a similar access-gating approach for its own cybersecurity model.

For a business owner, this is an early signal that the strongest AI tools will not all be available in the same way as general chat assistants. If your team works in software, IT, security, finance, or handles sensitive systems, expect more capability to come with more permissions, logging, review, and vendor checks. That is sensible, but it also means you need a clear view of which AI tools people are using, what data they can access, and who is accountable for them.

This is the kind of thing we build into an AI command centre, where powerful tools, access rules, and operating oversight sit in one practical view.