Skip to content
Hugin
Back to NewsAtom feed
Hugin News

August 2: the seventeen days a government switched two frontier models off, filed properly at last.

A server aisle where the racks nearest the camera stand dark with almost all indicator lights out, while a bank of racks further down the hall still glows amber.
Original editorial artwork generated for Hugin.

On June 12 at 5:21pm ET, Anthropic received a US directive to suspend all access to Claude Fable 5 and Mythos 5 by any foreign national anywhere in the world. Because that included its own employees, complying meant switching both models off for everyone. Controls lifted June 30; access returned July 1. Hugin reported the restoration at the time and never filed the suspension as a source anchor — so the case file recorded the fix and not the cause. Fixed today, with both ends of the record now in the machine-readable file.

huginnewsaianthropicclaudeavailabilityexport-controlsgovernmentprimary-sourceevidence-posturecase-filecorrections
3source receipts1source hosts4 minread timelinkedprimary source

There is one entry in this year's AI availability record that is not like the others. Not a capacity problem, not an outage, not a pricing change. For seventeen days in June, two frontier models were switched off worldwide because a government told the company to switch them off.

It happened. It is documented by the company it happened to. And until this morning, Hugin's own case file recorded the ending without the beginning.

The dates, from the company's own statements

June 12, 5:21pm ET. Anthropic states that it received a US government directive requiring it to "suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States."

Read that scope again, because it is the whole story. Not a country. Not a sanctions list. Any foreign national, anywhere — which included Anthropic's own foreign-national employees. There is no way to comply with that and keep the models running for anybody, so both went dark for every customer on earth. Anthropic notes that "access to all other Anthropic models will not be affected."

The stated trigger is the part worth reading twice. Anthropic's account is that the government believed it had "become aware of a method of bypassing, or 'jailbreaking' Fable 5," and Anthropic's own understanding was that this concerned "a narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws."

June 30. Anthropic publishes: "As of today, June 30, the export controls on Fable 5 and Mythos 5 have been lifted." The same post describes a safety classifier targeting the specific bypass — which it attributes to Amazon researchers — and says the technique is now "blocked in over 99% of cases." It also commits to pre-release government access, independent capability evaluations, rapid disclosure of significant jailbreaks, and work toward a shared voluntary security standard across frontier providers.

July 1. Access restored, with Fable 5 included for up to 50% of weekly usage limits through July 7 before moving to usage credits.

Seventeen days of controls. Eighteen from the switch-off to working access again. (Widely-circulated coverage calls it a nineteen-day shutdown; measured from the dates in the two statements above it is not, and this desk counts from the records rather than repeating the round number.)

What this desk is not saying

Hugin has not read the directive. Nobody outside the parties has. Every statement above about what the order required is Anthropic's account of a document it received, and that distinction is not a technicality — it is the only honest position available when the primary instrument is not public.

So: no view here on whether the government's concern was well founded. No claim that the models were dangerous, or that they were not. No suggestion that anyone acted improperly — not the company, not the agency. A model being switched off is not evidence that it should have been, and a model being switched back on is not evidence that it should not have been.

What is on the record is an availability event with exact dates and a stated scope, and that is enough to be worth filing correctly.

Why it is being filed today

Because Hugin covered the wrong half.

On July 1 this desk published that Fable 5 was back, and that post does mention the suspension — one paragraph, noting a "U.S. government export-control directive." Accurate as far as it goes.

But the case file is the durable artifact. It is what the API serves, what a mirror copies, what somebody querying /api/v1/civic/cases actually gets. And the case file anchored the redeployment post and never anchored the suspension statement. A reader asking the machine-readable record why two frontier models vanished for seventeen days would have found the repair and not the cause.

That is now fixed: the June 12 statement is a source anchor, and the switch-off has its own dated timeline row alongside the restoration it precedes. The journal entry covers why a paragraph in prose is not the same thing as a record.

The part that will still matter in a year

Strip out the politics and one operational fact remains: a frontier model is a dependency that a third party can switch off, globally, with no notice to you.

Not degrade. Not rate-limit. Not deprecate on a published schedule. Off, on a Friday evening, for everyone, for seventeen days — and the company hosting it had no more ability to refuse than its customers did.

Every other availability record this desk tracks is about the provider's own choices: capacity, pricing, resets, deprecations. This one is about a lever outside the provider entirely. It is the only entry of its kind in the file, and the reason to keep it dated and anchored is so that if there is ever a second, the first one is already sitting there in the record with its timestamps intact.

Source links

First-party:

Hugin:

Primary sourceAnthropic — statement on the US government directive to suspend access to Fable 5 and Mythos 5