The US government asks OpenAI to slow its next model’s release

At a glance:

• The Trump administration has requested OpenAI stagger the release of its upcoming frontier model.
• Access will be granted to a short list of trusted partners via a customer-by-customer approval process.
• The directive comes from the Office of the National Cyber Director and the Office of Science and Technology Policy.

A shift from voluntary to directed oversight

For years, the debate regarding the pace of artificial intelligence development has been confined to internal safety teams and external academic critics. However, a new development suggests that the era of voluntary safety commitments is transitioning into a period of direct government intervention. The Trump administration has formally requested that OpenAI stagger the rollout of its next major model, marking the first time the US government has preemptively asked an American AI firm to restrict a launch before it occurs.

OpenAI CEO Sam Altman communicated this development to staff on Wednesday, explaining that the government intends to oversee a preview period. During this window, access will not be open to the general public or all enterprise clients at once. Instead, the government will reportedly be "approving access customer by customer," ensuring that only a vetted list of trusted partners can interact with the new capabilities initially.

Cybersecurity concerns drive the mandate

The request does not appear to stem from a single regulatory office but rather from a coordinated effort between two key bodies: the Office of the National Cyber Director and the Office of Science and Technology Policy. Notably, the impetus for this restriction is framed around cybersecurity risks rather than concerns regarding market competition or the generation of harmful content. The primary worry is the potential for a sufficiently capable model to be weaponized if it falls into the wrong hands.

By implementing a staggered rollout, the administration aims to limit exposure during the model's most vulnerable initial phase. This approach mirrors a mechanism OpenAI previously utilized for GPT-5.4-Cyber, which was released exclusively to vetted security teams under a Trusted Access programme. This precedent suggests that the company is already accustomed to high-security, gated deployment models for its most sensitive iterations.

The implications for the AI arms race

This intervention marks a significant departure from the hands-off posture that has characterized US AI policy to date. Rather than relying on after-the-fact evaluations, the government is now actively shaping the release schedules of leading laboratories. This trend was signaled recently when rival Anthropic saw its most capable offerings pulled from the market under a government directive, suggesting a broader movement toward proactive control.

For OpenAI, the arrangement presents a complex strategic trade-off. On one hand, a staggered rollout slows the company's ability to monetize its newest technology and put it in front of a broad developer base—a critical factor in a market where competitors move with extreme speed. This comes shortly after the enterprise launch of GPT-5.5, making the timing particularly sensitive for OpenAI's commercial momentum.

On the other hand, the arrangement provides a layer of political insulation. If a model released under government supervision encounters unforeseen issues, OpenAI can point to the explicit involvement and approval of federal agencies as a form of shared responsibility. As the industry watches, the central question remains whether this customer-by-customer approval process will become the standard template for all future frontier model releases.