The Trump administration is moving the gate for frontier AI access, and it is not the labs deciding anymore. CNBC reported Friday, citing two people familiar with the matter, that the administration is dictating which companies and entities get access to models from Anthropic and OpenAI. Until now, the labs made those choices themselves.

This is a real power shift for executives trying to plan product roadmaps and partnerships. In other words, the bottleneck is no longer just technical capability or commercial terms with Anthropic or OpenAI. It is the question of who Washington lets in, and when.

CNBC’s report also points to how access used to work, at least in one lane. Anthropic controlled access to its Mythos cybersecurity model through an initiative called Project Glasswing. OpenAI, in parallel, ran an access program of its own, CNBC said, though the excerpt you provided cuts off after “OpenAI ran …”. Taken together, the implication is clear: the industry’s default model was “the lab curates access,” often with structured programs and specific onboarding pathways.

What changes when Washington gets involved is not just who gets approved. It is the incentives around compliance, risk, and speed. Labs already had to manage model safety concerns and manage demand. But if the White House is now deciding access recipients, those safety and governance discussions get pulled into a political decision channel. For boards and leadership teams, that changes how you interpret timelines, because approvals can become dependent on policy priorities rather than solely on a company’s technical readiness or stated use case.

There is also an organizational dynamic here that will not be lost on investors and operators. When the labs controlled access, negotiations and governance largely lived between the model providers and the customer companies, plus any external scrutiny that already exists in the AI policy ecosystem. Once access is controlled by the administration, the “deal” stretches outward. Companies that once focused on vendor diligence now need to understand how regulatory framing works across agencies and jurisdictions, and how policy shifts can alter what is possible even after a technical relationship is underway.

For cybersecurity and other sensitive domains, the Mythos example matters. CNBC notes Anthropic used Project Glasswing to control access to its Mythos cybersecurity model. Cybersecurity is exactly the kind of area where governments tend to care about who can use what, for what purpose, and under what constraints. If access decisions are moving toward political control, model deployment decisions will likely start looking more like regulatory compliance planning and less like pure commercialization.

The second-order effect for executives is that access becomes a strategic asset, not just a procurement input. Frontier models can improve products, accelerate research, and create competitive differentiation. If access is rationed or reprioritized through administration decisions, then partnerships and customer pipelines may hinge on factors unrelated to engineering alone. That could reshape how boards evaluate risk. “We can build it” stops being sufficient. Leadership teams also have to ask “will we be allowed to use it in the first place,” especially when the model use touches government priorities or regulated sectors.

Zoom out and it gets bigger. The report suggests a reallocation of power from frontier AI labs to the White House over “who gets access,” which is a governance lever. If this becomes a durable pattern, it could influence everything from which verticals see early deployments to how quickly firms can iterate based on model updates. For executives at companies considering collaboration with Anthropic or OpenAI, the strategic stake is simple: your AI strategy might now require a policy strategy as well.