Claude Fable 5 and Mythos 5 are back — but the export-control precedent survives

On 30 June the US Commerce Department withdrew the export controls it had placed on Anthropic's two newest models, Claude Fable 5 and Mythos 5, ending an eighteen-day standoff that began only three days after the models launched. The models are back. The licence requirement is gone. Global access is restored. But the episode is not really the story of two models being switched off and switched on again. It is the story of the mechanism used to do it — and that mechanism is still on the shelf, fully loaded, its precedent intact.

On 12 June, Commerce Secretary Howard Lutnick sent Anthropic a Bureau of Industry and Security "is-informed" letter ordering it to suspend access for any foreign national worldwide, citing reports of a jailbreak that could turn the models into unrestricted cyber tools. Because Anthropic cannot filter users by nationality in real time across global clouds, the only way to comply was a universal global shutdown — the models went dark for everyone. White House adviser David Sacks said Anthropic had refused to fix the flaw; Anthropic disputed both the severity of the jailbreak and the opacity of the process. What matters analytically is the category of the action, not the merits of either claim.

The sequence moved quickly, from directive to lawsuit to partial thaw to full rescission, and the compressed timeline is part of why the legal question was never resolved — the pressure that built the case also dissolved it.

Legion's complaint was unusually specific, and its arguments are the ones a court will eventually have to weigh. It contended that no existing export control covers hosted AI models or their outputs; that the directive exceeded the authority granted under ECRA §4817(b)(1), which requires formal rulemaking and allied coordination that "never occurred"; that AI outputs are protected informational materials under the Berman Amendment; and that the order was "materially underinclusive" because OpenAI's GPT-5.5 had comparable capabilities yet was left untouched. Taken together, these are not procedural quibbles — they go to whether the executive had any lawful basis to act at all.

The court never ruled. The government approved a partial reopening on 26–27 June, then rescinded the controls entirely on 30 June after Anthropic completed a national-security review with Commerce. A live controversy became a moot one. From a founder's perspective that is a clean outcome; from a precedent-watcher's perspective it is the worst kind — the power was exercised, challenged, and then quietly withdrawn before any judge had to say whether it was lawful. Anthropic co-founder Tom Brown reportedly led the negotiations with the administration, though that detail is reported rather than confirmed.

Read the settlement carefully and the shape of the new normal appears. The models return not because the government conceded it had overreached, but because Anthropic accepted a set of ongoing commitments — detection, standards-setting, reporting — and Commerce kept the right to "reevaluate." That is a negotiated licence to operate, not a vindication. The same federal gate has since appeared around OpenAI's government-gated model preview, which means Anthropic is no longer the isolated case; the instrument is now something any frontier lab can expect to meet.

So the durable question is the one Legion asked and no one answered: can the executive switch off a commercial hosted model under export law, with no rulemaking, no court and no Congress? Bipartisan lawmakers pressed for the legal basis — Representative Lori Trahan's summary was "No law. No process. No oversight." — and got no test case, because the test case evaporated. The controls on Fable 5 and Mythos 5 are gone. The theory that justified them is not. Until a court or Congress draws the line, the most consequential thing that happened this month is not the reversal — it is that the button was proven to work.