Fable 5 is back after 18 days. Don't celebrate the precedent.

Here's the relief narrative you're being sold: Washington threatened Anthropic's newest models, cooler heads prevailed, everyone's back online, move along. On 30 June 2026 the Commerce Department withdrew the export controls on Claude Fable 5 and Mythos 5, no export licence required, and Fable 5 was restored to global users across Claude, Claude.ai and Claude Code. Happy ending. Roll credits.

Don't. Look at what actually happened over 18 days, because the ending is the least interesting part. A cabinet secretary switched off a live commercial product for the entire planet with a letter — and then switched it back on once the company signed terms. That is the story. Everything else is theatre.

Eighteen days from global kill-switch to global restore. If you think that speed is reassuring, you have it backwards. The speed is the point: there was no process slow enough to be a process. A letter went out, a market vanished, a deal got signed, the market returned. No statute was amended. No judge ruled. Congress found out from the newspapers.

Anthropic didn't win an argument; it accepted terms. In Lutnick's letter, the company agreed to proactively detect and address security risks, to help the government write standards for upcoming models, and to inform the government of 'malicious activity.' In plain English: a private lab is now a standing reporting partner to Commerce on the models it hasn't even shipped yet. And Commerce, for its part, 'reserves the right to reevaluate' — which is bureaucrat for 'the switch is still on the wall.'

Legion LegalTech's suit was the interesting one, and the rollback quietly killed it. Legion argued something Washington really did not want litigated: that no existing export control covers hosted AI models or their outputs; that the directive exceeded ECRA §4817(b)(1), which requires rulemaking and allied coordination that 'never occurred'; and that AI outputs are protected informational materials under the Berman Amendment. It also called the order 'materially underinclusive' — OpenAI's GPT-5.5 had similar capabilities and wasn't touched. Selective enforcement, in other words.

You resolve a lawsuit like that two ways: a judge tests your legal theory in open court, or you make the injury disappear so the case goes moot. Guess which one Commerce picked. Lift the controls, Legion's 'irreparable harm' evaporates, the standing question with it — and the legal basis for the original letter never gets ruled on. Convenient. The bipartisan complaints didn't get an answer either; Rep. Lori Trahan's line — 'No law. No process. No oversight.' — is still just hanging there.

Two things can be true. The jailbreak concern wasn't nothing — a frontier model that can be talked into becoming an unrestricted cyber tool is a genuine problem, and 'do nothing' was never the responsible option. And Anthropic played a lousy hand well: handed a directive it couldn't technically comply with narrowly, it took the reputational hit of a global shutdown rather than pretend it could filter by passport, disputed the severity in public, and — per reports, so take it as reported — leaned on co-founder Tom Brown to negotiate the exit. That's competence under an impossible constraint, not capitulation.

But competence by the company doesn't launder the mechanism by the state. The lesson every frontier lab just learned is that your product's on-switch is, in a pinch, Commerce's on-switch — and the sister saga around OpenAI's government-gated GPT-5.6 preview shows the same federal gate is now swinging at more than one lab. Anthropic is no longer the singled-out one. That's not comfort. That's escalation. The controls are gone. The template is filed. Celebrate the reboot if you like; just don't mistake it for the all-clear.