12 June 2026: the week the US switched off Fable 5

4 min read · 749 words
Share:
Michael Darius Eastwood
Michael Darius Eastwood · Independent AI alignment researcher
Published
Michael Darius Eastwood · Timeline · 3 July 2026
Michael Darius Eastwood, independent researcher, London: originator of the embedded-correction alignment thesis (manuscript 8 December 2024, SHA-256 anchored: f0d1f38f).

On 12 June 2026, the US Department of Commerce and the Bureau of Industry and Security ordered Fable 5 and Mythos 5 offline. It was the first export-control action against a deployed commercial AI model in history. This article records what happened, what the register says about it, and what it very deliberately does not say.

Row 10. Institution: US Commerce and BIS. Event: Fable 5 and Mythos 5 ordered offline, first export-control action against a deployed commercial AI model. External date: 12 June 2026. Register class: CONVERGENT. Gap: approximately five months after the book. Sources: Harvard Law Review blog June 2026, CSA whitepaper, Reuters.

What the action was

The order took two production models offline for export-control review. The instrument was the same one BIS uses for advanced semiconductor equipment: an administrative order backed by the enforcement architecture of the export-control regime, not a legislative kill switch. That distinction matters for what the register credits and what it does not.

Why it is in the register

The January 2026 book, at the Chokepoint sections, argues that the AI supply chain concentrates leverage in specific choke points and that the state that controls those choke points controls the deployment envelope. The book does not predict Fable 5 by name. It predicts, in prose, that a first-mover state action of exactly this class would happen at exactly these choke points. Five months later, an action of exactly that class happened. Priority on the enforcement belongs to BIS. Priority on the analysis belongs to the many trade-policy scholars whose work the book cites. The register credits the direction, and only the direction.

What the register does not say

It does not describe the action as a kill switch. That framing is forbidden by the canonical facts file and blocked by the truth gate. A kill switch is a device inside a system; an export-control order is a regulatory instrument against the sale or operation of that system. The Chokepoint framing is about administrative leverage, not about deadman-switch hardware. The register keeps the two categories separate because conflating them would mislead readers about what is technical, what is political and what has changed.

Row 12: the EU response, one week earlier

Nine days before the US action, on 3 June 2026, the European Commission published the EU Tech Sovereignty Package, comprising Chips Act 2.0 and CADA. Commissioner Virkkunen said: "We want to be sure nobody has a kill switch." The register lists this as row 12 and instructs strictly that it be described as anti-foreign-kill-switch sovereignty framing, not as the EU introducing a kill switch. This is the framing rule the truth gate enforces, and this article observes it.

Row 11: Macron at G7 Évian

Seven days after the US action, on 19 June 2026, at G7 Évian, President Macron delivered a "strictly nationalist" critique and proposed a "trusted partners" framework, with the line: "nobody will buy US AI if they fear it can be switched off overnight." The register lists this as row 11. Together, rows 10, 11 and 12 land inside seventeen days: an EU sovereignty package, a US enforcement action, and a G7 critique. The register calls this the sovereignty cascade and dedicates one row to each event, with primary sources on each row.

The honest caveat

The book made the analytical case in January 2026 that this class of action would happen. The book did not predict, and does not claim to have predicted, the specific timing, the specific models, or the specific choice of BIS as the enforcing agency. The register is precise about that difference, because the register is what it is, and it would rather be smaller and true than larger and wrong.

From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.

Buy on Amazon UK Amazon US

Stay informed

New posts on AI alignment, convergence evidence, and the ARC/Eden research programme.

Get updates →