On June 12, the Commerce Department ordered Anthropic to suspend all access to its most powerful AI model, Mythos 5 — including access by the company’s own employees who happened to hold foreign passports. On June 26, Commerce Secretary Howard Lutnick signed a letter restoring access to Mythos 5 for roughly 100 “trusted partners,” a group that Reuters reports includes Fortune 500 companies and federal agencies.

Fourteen days. That’s how long it took the U.S. government to go from “this technology is too dangerous for anyone to touch” to “here’s your VIP pass.”

Forget, for a moment, the debate about whether Mythos 5 is actually dangerous. Forget the argument about whether Anthropic is a safety-conscious lab or a regulatory-capture operation with better branding. Those are the conversations the industry is having. They’re missing the thing that happened in plain sight: the government just demonstrated that it can flip the switch on frontier AI access at the tempo of a cable-news decision cycle, and nobody in a position of power seems to find that remarkable.

Two Weeks Is Not Deliberation

Let’s be clear about what happened in that fourteen-day window. The initial order — the one that locked down Mythos and its sibling model Fable — was broad enough to bar foreign nationals working at Anthropic from accessing their own company’s systems. That’s not a surgical intervention. That’s a tourniquet.

Then, according to Lutnick’s letter, Anthropic “worked with the U.S. government to address risks associated with the Covered Models.” In two weeks. During a period when, by the government’s own logic, the models were so sensitive that even the people who built them couldn’t look at them if they held the wrong passport.

What kind of risk assessment gets done in two weeks? What kind of vetting process produces a list of 100 trusted institutions in the time it takes to get a passport renewed? The answer, almost certainly, is: not a thorough one. The list was assembled from whoever was already in the room. The speed makes that clear.

“We were in the middle of a compliance scramble, and then suddenly the phone rings and it’s a different person from Commerce asking if we want to be on the early-access list,” said one engineer at a company that made the cut, in a Slack message reviewed by this columnist. “I don’t think anyone on our side understood the criteria.”

The Precedent Nobody’s Naming

What the Commerce Department just established is a framework in which the executive branch can, with a letter, shut down access to a general-purpose technology — and then, with another letter, selectively restore it to a favored list. The legal basis is export-control authority. The practical effect is a licensing regime for intelligence.

Imagine this applied to any other general-purpose technology. Imagine the Department of Energy blocking access to a new battery chemistry, then two weeks later granting it to 100 “trusted” companies. The antitrust bar would be on fire. The trade press would run a thousand hot takes. The business press would be asking hard questions about who got on the list and why.

With AI, the reaction has been muted. The safety community is relieved that someone is doing something. The industry is relieved that the something includes them. The only people asking whether this is a good idea are a handful of civil-liberties groups and the usual skeptics on social media, and they’re mostly asking the wrong question — about the model’s capabilities, not about the governance mechanism being built in real time.

The Real Risk Is the Process

There’s a version of this story where the government moves deliberately. It issues a preliminary order, conducts a months-long review, publishes criteria for trusted-partner status, solicits public comment, and then — grudgingly, transparently — begins granting access under a framework that Congress has debated and the courts can review.

That’s not what happened. What happened was two letters, fourteen days apart, from a Commerce Secretary who was confirmed less than a year ago. The letters cite no statutory framework beyond export-control law. The list of trusted partners is not public. The criteria for getting on it are not public. The safeguards that satisfied Lutnick are not public.

This is not a governance regime. It’s a procurement process with a security clearance.

And it will be repeated. The next time a frontier model triggers a panic — and there will be a next time — the playbook is now written: shut it down, pick up the phone, and decide who gets to be in the club. The speed of the Mythos 5 reversal suggests that the people running this process don’t see it as a dangerous precedent. They see it as a feature.

The Comfortable Silence

Why isn’t this a bigger story? Partly because the companies that would normally scream about government overreach are on the list. Partly because the national-security argument has a powerful gravitational pull — nobody wants to be the person defending the right of foreign adversaries to access frontier AI. And partly because the AI industry has spent years telling policymakers that these models are existentially dangerous, and now it’s getting the regulatory attention it asked for.

But a regulatory regime built on panic and speed is not the same thing as a regulatory regime built on law and deliberation. The former is what you get when you treat every new model release as a potential national-security event. The latter is what you get when you actually govern.

Fourteen days. That’s all it took. The model may be safe. The process isn’t.

Sources