The Trump administration asked OpenAI to limit the rollout of GPT-5.6 to a small group of government-approved partners before a broader public release. CEO Sam Altman informed employees during a company-wide Q&A that the model would launch as a "limited preview" with federal officials approving each enterprise customer individually.
The request came from the Office of the National Cyber Director and the Office of Science and Technology Policy. Commerce Secretary Howard Lutnick advised against any launch without cross-agency approvals. OpenAI complied but made clear the arrangement is "not our preferred long-term model."
The why matters more than the what.
The cybersecurity trigger
Both OpenAI and the administration view GPT-5.6 as "on par" with Anthropic's Mythos in advanced cybersecurity capabilities. That alignment triggered the request. The government is concerned the model could be used to identify software vulnerabilities at scale without human oversight.
The concern is real. Frontier models with advanced reasoning can identify unknown security flaws in code. That capability is valuable for defense. It is also dangerous if distributed without controls. The administration chose to apply controls.
The Anthropic precedent
Three weeks earlier, the Trump administration ordered Anthropic to take Fable 5 and Mythos 5 entirely offline, citing national security risks from a potential jailbreak. Anthropic complied. The precedent was set: the government can shut down models unilaterally.
OpenAI's proactive negotiation with the administration since before the Anthropic incident suggests the company learned that lesson. Rather than face shutdown, OpenAI agreed to a controlled rollout. It is a calculated surrender with better terms than Anthropic received.
The new era of pre-release approval
This is the first time the US government has actively controlled access to a frontier AI model before public release. Previous interventions focused on post-launch oversight or voluntary commitments. This is different. This is pre-release vetting on a customer-by-customer basis.
The implication is significant. AI companies no longer control their own launch timelines. The government does. A model can be technically ready but politically blocked. A company can have paying customers waiting but be unable to serve them without federal approval.
Altman framed the arrangement as collaborative. The practical reality is different. OpenAI does not decide when GPT-5.6 launches. Washington does.
The competitive landscape
While OpenAI navigates federal restrictions, Google and Meta continue shipping frontier models with minimal government interference. That asymmetry could shift enterprise customers toward alternatives if the vetting bottleneck drags on.
The administration's intervention sends a signal: frontier AI will be treated as critical infrastructure subject to federal gatekeeping. Companies that negotiate proactively fare better than companies that resist. But all of them now need federal relations as part of product development.
The era of self-regulated AI deployment has ended.
