Update, 08:20 BST: Anthropic has now confirmed the models are actually back. Fable 5 and Mythos 5 are live again globally as of today — across Claude.ai, the API, Claude Code and Claude Cowork, with the cloud partners (AWS, Google Cloud, Microsoft Foundry) being re-enabled as fast as possible. On paid plans, Anthropic says Fable 5 is included for up to half of your weekly usage limit through 7 July, and available on usage credits after that. The company also detailed how the impasse was resolved: it trained a new cybersecurity classifier it says blocks the specific jailbreak technique — the one a partner, reported to be Amazon, had flagged — in more than 99% of cases, and noted its own testing found that less capable models, including Opus 4.8 and GPT-5.5, could reproduce the same findings. In other words, the vulnerability wasn't unique to Fable 5 — which was Anthropic's argument from the start. Our original report follows.

Good morning. The held breath we described yesterday has been let out. After eighteen days in the dark, the US government has lifted the export controls on Anthropic's two most powerful models — Fable 5 and Mythos 5 — and the company says it will begin switching them back on. The single biggest story in AI this month has an ending. It's worth reading the terms of it closely, because the terms are the story.

The ban is over

Late on 30 June, Anthropic said it had received notice that the Department of Commerce had lifted the export-control directive imposed on 12 June — the order that forced it to disable Fable 5 and Mythos 5 worldwide, for every foreign national, including its own staff. "We'll begin restoring access tomorrow," the company wrote, "and will share an update soon." By its own timeline, that puts the models coming back from today, 1 July.

The reason the ban was liftable is a deal. In a letter to Anthropic, Commerce Secretary Howard Lutnick said the company no longer needed an export licence because it had agreed to three things: to proactively detect and address security risks in the models, to work with the government on standards for future releases, and to report malicious activity to the authorities. That's not a return to how things were on 11 June. It's a negotiated security compact — a frontier lab shipping its best models under commitments made to Washington.

Why the terms matter more than the timing

Yesterday we argued that the rhythm of AI has quietly shifted from releases to permissions — that the most powerful force in the field right now isn't a model, it's a queue of government decisions. Today doesn't overturn that; it confirms it. Fable 5 is coming back not because a ban expired or a court intervened, but because a permission was granted, in exchange for a set of ongoing obligations. The gate opened — but there is now, unmistakably, a gate, and a keeper, and a price of passage. That is the durable news here, well beyond this one model.

A few honest caveats, because the aggregators are already overshooting. Anthropic said it would begin restoring access — deliberately hedged language that doesn't promise everything is live everywhere today. And while Commerce has lifted its control, we could not confirm that the specific Pentagon and NSA sign-off described last week was the mechanism; what's on record is that Commerce removed the licence requirement after the security commitments were made. No fees or identity-check conditions were disclosed beyond those three commitments. So: the ban is genuinely over, access is being restored from today, and the fine print is a compact, not a fee.

The story it quietly closes

The lifting also overtakes a subplot. Austria's letter urging the European Commission to explore hosting Anthropic inside the EU — floated on Sunday precisely because Europe had been cut off — is still, as of this morning, unanswered by Brussels. It may now simply be moot: the access Vienna was worried about losing is being restored before the Commission has even replied. Europe's sovereignty reflex was real, but it was overtaken by a US–Anthropic deal in a matter of days.

One thing that has not changed: the other frontier models remain behind their own gates. OpenAI's GPT-5.6 "Sol" is still limited to about twenty government-approved partners, with general availability promised only "in the coming weeks," and Google's Gemini 3.5 Pro is still delayed into July. The gate swung open for Anthropic today. It's worth remembering how many models are still on the other side of it.

This has been On The Wire.