OpenAI has launched a limited preview of GPT-5.6, a new model the San Francisco-based company says carries powerful cybersecurity capabilities, with access gated to users who have been vetted by the US government. The move marks a sharp departure from OpenAI's typical public rollout playbook and signals the company is treating this release as sensitive national-security infrastructure rather than a consumer product.
Government Vetting as the New Moat
By conditioning access on US government clearance, OpenAI is effectively drawing a line between what it considers general-purpose AI and something closer to dual-use technology. That distinction matters commercially: it positions GPT-5.6 as a procurement-grade product rather than a subscription tier, routing it into a spending category where contracts are larger, cycles are longer, and switching costs favor incumbents. Competitors without comparable government relationships face a structural disadvantage in bidding for the same work, regardless of their models' underlying performance.
The vetting requirement also shifts liability. If a model with strong cybersecurity capabilities ends up misused, the paper trail runs through a government screening process — not just OpenAI's own terms of service.
What "Powerful Cybersecurity Capabilities" Actually Means for the Market
OpenAI's description of the new models as having powerful cybersecurity capabilities is the detail that explains the access controls. Cybersecurity AI is a genuine dual-use problem: the same capability that helps a defender identify a vulnerability can help an attacker exploit one. That tension is precisely why controlled distribution makes commercial sense here, separate from any regulatory pressure.
For enterprise buyers in defense, intelligence contracting, and critical infrastructure, a government-vetted model clears procurement hurdles that a standard commercial release cannot. The question is whether OpenAI eventually expands the preview to a broader set of credentialed organizations, or keeps GPT-5.6 in a separate, tightly managed product lane altogether.
Limited Preview, Unlimited Implications
OpenAI is calling this a limited preview, which means the commercial terms, pricing structure, and eventual general availability timeline remain unannounced. What is clear is that the company is not treating GPT-5.6 as a feature update to an existing product — the restricted rollout and government-vetting requirement suggest OpenAI views the cybersecurity capabilities as categorically different from what it has shipped before, and is managing the release accordingly.