Three Months Ago, Forever

June 13, 22:40. US Secretary of Defense Pete Hegseth, posting to 603,000 views: "Three months ago, @DeptofWar kicked @AnthropicAI out of our building, forever. Every passing day proves why that was the right move."

Four Months, Not One Week

The published piece documented the June 12 directive as a collision between two regulatory logics, one European, one American, with EU users caught in both. That collision is real. But the directive itself was never primarily about either of those things. It was the latest move in a four-month domestic argument over who gets to set the safety limits on America's most capable AI, fought in public between a cabinet department and a single company.

Every non-US user, including everyone in the EU, inherited the outcome of an argument they were never party to, about a question, should the military get AI without human oversight on lethal force and domestic surveillance, that they were never asked.

Visible at Every Step

There is something genuinely new in how this is happening. Past technology transitions, nuclear, early cryptography export controls, played out mostly behind classification. This one is happening in the feed. A Secretary of Defense gloating in real time, a CEO's statement, a court injunction, an export directive, all visible, all timestamped, all four months apart and all colliding in one week. The institutions are improvising in public, and you can watch an administration sign a memo to accelerate AI on a Friday and disable the most advanced AI on the planet the following Friday.

Built It, Can't Use It

One detail worth sitting with: the foreign-national restriction in the June 12 directive lands on several of the people whose careers built the safety research the directive claims to be protecting. People who spent years on the question of how to make these systems safe now cannot access the systems they helped make safe, because of where they were born. Whatever the directive is actually targeting, it does not appear to distinguish between that and this.

What This Adds

The published piece found that the EU holds roughly 5% of global AI compute and regulates products it does not produce at the frontier. This adds a sharper edge to that finding. The mismatch is not only that some actors lack productive capacity. It is that even the country with the capacity is visibly struggling to build governance at the speed the technology moves, producing contradictory policy within the same administration inside a single week.

This is not a one-off. The Anthropic/Pentagon relationship has been an open, escalating, publicly fought argument since February, and the June 12 directive reads less like a new event than like the next scheduled instalment. Expect more.