Good morning. Some mornings the AI news is a flurry of launches; this is not one of them. Tuesday opens on a holding pattern — and the holding pattern has quietly become the story. Three weeks into the most consequential stretch of AI policy this year, the most powerful force in the field isn't a model. It's a queue of decisions nobody has made yet.

Day eighteen, and Fable 5 is still dark

Anthropic's most capable model, Fable 5, has now been offline for eighteen days — switched off on 12 June by a US export-control order, along with the cyber-focused Mythos 5. Last week brought the first thaw: on 26 June, Commerce Secretary Howard Lutnick's office cleared Mythos 5 for roughly a hundred vetted US critical-infrastructure organisations. Fable got no such reprieve.

The expectation, reported by Axios on 27 June and echoed by Forbes on 29 June, is that Fable's return is close — "this week," on anonymous sourcing, with Pentagon and NSA sign-off described as the last hurdle. But expectation is not confirmation. As of this morning there has been no announcement that Fable 5 is back, and a crop of aggregator headlines already declaring it "restored" are running ahead of the facts. The honest status is the one we gave it yesterday and the day before: expected, not returned. When it actually switches back on, that will be the news. It hasn't yet.

Austria's letter is still sitting in an inbox

Sunday's genuinely new move — Austria urging the European Commission to explore hosting Anthropic inside the EU rather than keep depending on access Washington can revoke — has so far been met with silence. Commission executive vice-president Henna Virkkunen's office has not responded publicly as of this morning; no other government has echoed it. As we laid out yesterday, the idea runs straight into the wall of US export jurisdiction, which follows the company and its technology regardless of address — so the interesting question now is whether Brussels treats the letter as a serious opening or a non-starter. That answer, too, is pending.

Everywhere you look, a "coming soon"

The pattern repeats across the board. OpenAI's GPT-5.6 "Sol" remains locked to roughly twenty government-approved partners, with general availability still promised only as "the coming weeks" — and still no independent benchmarks, only OpenAI's own figures. Google's Gemini 3.5 Pro, expected in June, has reportedly slipped to July and stays in limited enterprise preview. Two of the three biggest US labs have their flagship offerings behind a gate or a delay, and the third is waiting on the Pentagon.

It is worth naming what that adds up to, because it's easy to miss when nothing "happens." For most of the past two years the rhythm of AI was set by releases — each lab racing to ship the next model. Right now the rhythm is set by permissions: government clearances, export reviews, partner approvals, an unanswered diplomatic letter. The technology is, for the moment, ready and waiting; what everyone is actually watching is a set of decisions sitting with officials in Washington, and now Brussels.

That's the real state of play this Tuesday. Not a quiet day so much as a held breath. We'll be here when someone exhales.

This has been On The Wire.