The US Government Pulled the Plug on Its Best AI. Days Later, It Quietly Plugged It Back In.

In the span of a few days, the US government told one of the best AI labs on the planet not to ship its top models abroad, then changed its mind and let them go. The target wasn’t a foreign competitor. It was Anthropic, a San Francisco company. The whole thing was short, but it revealed something worth staring at.

What Actually Happened

It started with the Department of Commerce slapping export controls on Anthropic’s flagship models, Fable 5 and Mythos 5. In plain terms, the government put the brakes on serving or providing these models outside the country.

Here’s the wrinkle: this isn’t a physical good like a semiconductor. It’s software — a service you reach through an API. Commerce treated it as an “export” anyway, presumably under the familiar banners of national security and technological advantage.

Then came the reversal. Commerce withdrew the controls almost immediately. A move it had pushed hard, walked back on its own within days.

The Internet Called It a Light Switch

There wasn’t much online chatter about this one — which itself tells you how fast it flared up and fizzled out. But a few reactions caught the mood.

A Chinese-language YouTube channel framed it bluntly in a June 26 video: with a single word from the White House, “the plug got pulled on AI worldwide.” The clip cleared 16,000 views and over 1,000 likes. The point being that a service could flicker on and off at a government’s say-so.

A weekly tech roundup went further and put “Anthropic AI Shutdown” right in the title. Regulation was being read, correctly, as a service outage.

The interesting part is that the calls to lift the ban came first. Security expert Alex Stamos argued in a June 19 policy interview that the US should drop the export restrictions on Fable and Mythos. In the short window between the rule landing and getting yanked, the experts had already flagged it as an own goal.

Why Block It, and Why Undo It

The rationale writes itself. You don’t want your best-performing models landing in the hands of rivals or adversaries. Now that AI sits at the center of national competitiveness, “we keep the best weapon for ourselves” feels like an obvious instinct.

The instinct has three holes.

First, enforceability. You can’t inspect a model at the border. Chip-making equipment moves in containers through ports; a model does not. And plenty of competing models already match the performance. Block one, and there’s a substitute waiting.

Second, shooting yourself in the foot. If the US government stops US companies from serving their best models, that market gets filled by someone else’s models. That was Stamos’s core point: the security upside of blocking is smaller than the cost of handing over the market and blunting your own industry’s edge.

Third, the trust tax. What company confidently adopts a service that’s live today and dark tomorrow? The moment regulation starts flip-flopping, the whole AI ecosystem in that country loses its predictability.

The Real Story Is the Switch, Not the Tech

What grabs me here isn’t the substance of the rule. It’s the mechanism.

Old-school export controls were about objects. You name a specific piece of equipment, a specific chip, and you stop it at customs. There’s a process, it takes time, and that makes it predictable.

AI models break that model. A service running in the cloud goes on and off instantly with a single policy decision. That’s exactly why “the plug got pulled” was the phrase people reached for. This episode proved, in practice, that a government can gate access to the best-performing AI in the world with one switch.

The days-long reversal is a relief. It’s also an unsettling signal. The government showed it can flip that switch both ways. And a switch that’s been pressed once can be pressed again, anytime.

The Takeaway

This flare-up ended fast, but the question it leaves is heavy. The more AI becomes a national asset, the more we’re living in an era where a single government finger can decide whether a service stays on.

You can’t dismiss regulation outright — powerful technology needs oversight. But if a measure can be reversed within days, how carefully was it weighed to begin with? A government controlling access to the best AI like a light switch — do you read that as a safeguard, or as a new kind of risk?