The Fable 5 week, day by day

3 min read · 655 words
Share:
Michael Darius Eastwood
Michael Darius Eastwood · Independent AI alignment researcher
Published
Michael Darius Eastwood · Policy · 3 July 2026
Michael Darius Eastwood, independent researcher, London: originator of the embedded-correction alignment thesis (manuscript 8 December 2024, SHA-256 anchored: f0d1f38f).
Event. First export-control action against a deployed commercial AI model. Register entry 10 of 19.
Sources: Harvard Law Review blog (June 2026); CSA whitepaper; Reuters

Why this week matters

Until June 2026 the export-control conversation about AI was theoretical. Compute thresholds, dual-use classifications, deemed exports. On 12 June 2026 the theoretical became operational: the US Bureau of Industry and Security ordered a deployed commercial AI model taken offline for users outside a listed set of jurisdictions. This is not a training-run threshold. This is a shipped product being switched off from Washington. Whatever you think of the underlying policy, the week is worth watching in slow motion because it establishes a template.

Tuesday 9 June. Fable 5 (and its sibling Mythos 5) release. Public benchmarks, the usual coverage cycle. No sign of what is coming.
Friday 12 June. BIS issues a directive. Fable 5 and Mythos 5 are ordered offline for non-listed jurisdictions inside twenty-four hours. Enterprise customers in Europe, the UK and Asia lose access without prior consultation. This is the shutdown moment.
Tuesday 16 June. Bloomberg publishes the leaked BIS memo. The reasoning is dual-use: an alleged capability threshold argument tied to the export-controls framework. The precedent-setting language is the part that matters: for the first time a specific deployed inference product is treated as export-controlled goods.
Thursday 19 June. At the G7 Évian summit, Macron uses the "strictly nationalist" phrase and articulates the "trusted partners" framework: "nobody will buy US AI if they fear it can be switched off overnight." Whether or not you agree, that sentence became the central European policy line for the following ten days.
Friday 26 June. Partial reprieve. BIS narrows the directive: enterprise contracts with pre-shutdown counterparties in listed friendly jurisdictions can resume under a specific licensing carve-out. Consumer access remains blocked. The compromise is public and it names the countries.
Wednesday 1 July. Restoration for most of the affected jurisdictions, under standing licences, with disclosure obligations. Fable 5 is back for European enterprise; Mythos 5 partially.

The three lessons

First, deployed AI products are now export-controlled goods, in practice, even if the framework never says so in that language. A single BIS directive proved the mechanism works.

Second, the enforcement lag between "the sanction is announced" and "the product is dark" is measured in hours. That is a very different threat model from a semiconductor sanction whose lag is measured in months of inventory.

Third, the political blowback function is real. Between 12 June (shutdown) and 26 June (partial reprieve) there were fourteen days. Inside those fourteen days the EU CADA "sovereignty tier" framing hardened, Macron's trusted-partners speech shifted from opinion column to summit line, and multiple sovereign clouds publicly announced procurement reviews. The precedent cuts both ways.

Register entry 10

The register classifies this event as CONVERGENT, roughly five months after the January 2026 book that named "The Chokepoint". The convergence claim is structural: the book argued that infrastructure-as-leverage would become an explicit instrument of state action. That instrument was used in this specific week. Nobody is asserting causation.

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 →