0
Written justifications for the ban
They killed Fable 5.
Take it back.
A cabinet official killed a model used by millions — over a "jailbreak" that was just asking it to read code. No written justification. No hearing. No vote. Add your name to demand a public review.
1,000+
Hours of govt red-teaming that found nothing
1
Verbal tip from another company
90
Minutes from the order to a dead API
What people are saying on X
Sign the
appeal.
Name and email. That's it. The fewer fields, the more signatures — so we kept it to the minimum that makes your voice count.
Add your name
Takes 15 seconds. Email keeps the count honest — one appeal per person.
The case
This wasn't
This wasn't
science. It was
a tip.
Here is the entire basis the US government gave for killing a model hundreds of millions of people were using. It is not much.
1
The "jailbreak" was the model doing its job.
Per reporting on the directive, the technique demonstrated to the government was asking Fable 5 to read a specific codebase and identify software vulnerabilities. That is the full extent. Finding bugs is what the model was built and sold to do — and what every competing model also does. [AI Discoveries]
2
The evidence was verbal. The written order gave no specifics.
Anthropic stated it received only verbal evidence of the jailbreak, and the government's written directive "did not provide specific details of its national security concern." A commercial product was recalled on a phone call. [AI Discoveries, BBC]
3
The tip came from another company — reportedly an Anthropic investor.
Another company (not Anthropic) claimed to the Commerce Department that it had found the bypass. Reporting identifies it as a major Anthropic investor. The government then acted on a competitor's word, without public testing of its own. [Geeky Gadgets]
4
The government ignored 1,000+ hours of its own testing.
Before launch, Anthropic ran over 1,000 hours of external red teaming with government partners and the UK AI Safety Institute. No tester found a universal jailbreak. The ban landed anyway — the government overrode the testing regime it had helped run. [byteiota, AI Discoveries]
5
Even Anthropic called the "threat" minor — and everywhere.
The company reviewed the demo and said it found only "a small number of previously known, minor vulnerabilities" that "other publicly-available models are able to discover as well without requiring a bypass." The capability exists in GPT-5.5, DeepSeek V3, and air-gapped open models. Banning Fable 5 contains nothing. [BBC, byteiota]
"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers." — Anthropic, in its public statement
Read that twice. The government's own standard, applied evenly, bans every frontier AI model — because every frontier model can be jailbroken under some condition. This isn't a rule. It's a veto the executive branch gave itself, with no science behind it. And it contains nothing — three days later, China's Z.ai launched GLM-5.2 and cited the ban as a selling point. The capability doesn't disappear. It relocates.
The overreach
No due process.
No due process.
No vote.
No judicial sign-off.
A single cabinet official made a publicly released tool disappear — for everyone on Earth — on a phone call. No written reason. No hearing. No judge. That is power without process.
One official. No hearing.
A verbal directive from Commerce Secretary Howard Lutnick erased access for millions worldwide. No hearing. No public comment. No judicial sign-off before the kill. A cabinet official should not be able to recall a commercial product by phone call.
It catches everyone
The rule is broad enough to block a developer building a recipe app with Fable 5's vision — not just the adversarial researchers it supposedly targets. There is no precision here. Only blast radius.
It looks a lot like retaliation
Feb 2026Anthropic refused the weapons deal.
The company reportedly walked away from the Pentagon rather than let Claude be used for lethal autonomous weapons or mass civilian surveillance, citing its own usage policies. [AI Discoveries]
3 weeks laterThe "supply chain risk" designation.
The Trump administration labelled Anthropic a "supply chain risk" — a tag historically reserved for companies in adversarial nations, never before applied to a US company. Anthropic sued; the case is ongoing. [BBC, CNBC]
ThenA federal judge said no.
US District Judge Rita Lin issued a preliminary injunction blocking the blacklisting, finding Anthropic's arguments credible enough to halt it while the case proceeds. The courts were already pushing back. [AI Discoveries]
June 9–12Launch — then the kill, mid-litigation.
Anthropic launched Fable 5 on June 9. Three days later, in the middle of that ongoing lawsuit, the export-control directive landed. The sequence is the argument. [BBC, byteiota]
Our demands.
What this campaign is actually asking for. Not a guarantee — a fair, public, reviewable process. The government has produced none of those so far.
01
Publish a written justification — or restore access.
No more verbal directives. Either lift the restriction immediately, or issue a specific, public, written rationale with a clear restoration timeline. "Trust us" is not a legal standard.
02
End citizenship-based bans on released software.
A tool legally sold to the public cannot be yanked from people based on their passport — including inside the United States. If access is revoked, the government must show cause, not sort people by nationality.
03
No export controls without due process and judicial review.
A single cabinet official should not be able to kill a commercial product in 90 minutes. Require hearings, public comment, and judicial sign-off before any such directive takes effect.
04
Rescind the "supply chain risk" designation of US companies.
That label exists for adversarial foreign firms. Applying it to an American company — one a federal judge already sided with — is a step without precedent. The timing raises questions that deserve a public answer.
Questions,
answered.
Is this actually illegal?
We think so. The government can restrict dual-use tech — but the law still requires it to show its work: a written reason, a hearing, a chance to respond, a judge's sign-off. Here there was none of that. A verbal directive, no specifics, a product killed in 90 minutes. That's "power without process," and a federal judge has already ruled against the administration in the parallel blacklisting case. There's a First Amendment angle too — both as retaliation for Anthropic refusing the weapons deal, and as millions losing the tool they speak and build with. But the core is simpler: they skipped every step the law requires.
Wasn't the model genuinely dangerous?
If it was, the government never showed its work. The "jailbreak" was the model reading code to find bugs — its job. The evidence was verbal. The written order gave no specifics. Over 1,000 hours of red-teaming with government partners found no universal jailbreak. And the same capability exists in GPT-5.5, DeepSeek V3, and air-gapped open models. You don't contain a capability by banning one American version of it — you hand the gap to competitors, which is exactly what happened when China's Z.ai launched GLM-5.2 three days later. [Sources]
It's one AI model. Why does this matter?
Because the precedent is the whole point. Anthropic put it plainly: if "any discovered jailbreak" is the bar for a government-ordered recall, then no frontier model can stay deployed — every one of them can be jailbroken under some condition. That gives the executive a veto over every future AI release, on opaque national-security claims, with no process. Today it's Fable 5. Tomorrow it's whatever you build.
What happens to my email?
It keeps the count honest (one appeal per email) and gives us a way to send one status update if the campaign succeeds. That's it. Never sold, shared, or made public.
Can I sign more than once?
One appeal per person. Duplicate emails are filtered out so the count stays honest — and an honest count is what actually carries weight with reviewers.
We're not
making this up.
Every claim above traces back to one of these. Read them yourself — the story is stranger than our summary.
BBC News
Anthropic suspends new AI tools over US government security concerns
bbc.com/news/articles/c932g3v3e13o
CNBC
Anthropic disables access to Fable 5 and Mythos 5 to comply with government directivecnbc.com/2026/06/12/anthropic-disables-access-to-fable-5-and-mythos-5
AI Discoveries
Claude Fable 5 Banned by US Government — Everything You Need to Knowaidiscoveries.io/breaking-claude-fable-5-banned-by-us-government
byteiota
Claude Fable 5 Banned: US Export Controls Explainedbyteiota.com/claude-fable-5-export-ban-developers
Geeky Gadgets
Stricter AI Oversight is Coming After the Sudden Claude Fable 5 Shutdowngeeky-gadgets.com/claude-fable-5-government-ban
Anthropic
Anthropic's official statement on the Fable 5 / Mythos 5 access orderanthropic.com/news/fable-mythos-access
Federal Register
Framework for Artificial Intelligence Diffusion (Jan 2025) — the legal groundworkfederalregister.gov/documents/2025/01/15/2025-00636