On 12 June 2026, by order of the US government, Anthropic cut off access to its Fable 5 and Mythos 5 models for all of its customers. The stated reason: a flaw the vendor considers minor. For businesses, the real lesson lies elsewhere: an AI model can stop within hours, for reasons you do not control.
On 12 June 2026 at 5:21 p.m. (Eastern Time), Anthropic received an export-control directive from the US government, in the name of national security. The order was sweeping: suspend all access to Fable 5 and Mythos 5 for any foreign national, anywhere in the world, including Anthropic's own non-US employees. To stay compliant, the company had no choice but to disable both models for every customer. Other models remain available.
The letter did not spell out the precise nature of the concern. According to Anthropic, the government believes it became aware of a method to bypass, or "jailbreak", Fable 5.
Anthropic says it reviewed a demonstration of the technique. Its finding: it surfaces only a small number of minor, already known vulnerabilities, fairly simple ones that other public models can find without any bypass at all. The evidence provided amounts, in essence, to asking the model to read a codebase and fix its flaws. That is a capability offered by other providers (Anthropic cites OpenAI's GPT-5.5) and used every day by the defenders who keep systems safe.
A non-universal jailbreak only elicits sensitive information in narrow, specific cases. A universal jailbreak broadly bypasses a model's safeguards. To date, and according to Anthropic, no tester has found a universal jailbreak on Fable. The rare findings stay narrow and give an attacker no real uplift.
Anthropic stands by the position it set out at Fable's launch: safeguards so strict that some users find them too broad. Before release, the model went through thousands of hours of penetration testing with the US government, the UK AISI, third-party organizations and internal teams.
The vendor's conclusion: framed this way, Fable's risk sits at the level of models already in service across the industry.
Anthropic complies and removes access. Yet it disputes the principle: that a narrow and merely potential flaw should be enough to recall a commercial model used by hundreds of millions of people. Applied across the sector, that test would, in its view, freeze the deployment of every new frontier model.
Anthropic says it supports a government's right to block an unsafe deployment, but through a transparent, fair, clear and technically grounded statutory process. In its view, this directive does not meet those conditions. The company apologizes to its customers, calls it a misunderstanding and says it is working to restore access.
Whether you side with the government or the vendor, the takeaway for a leader is the same: an AI model can vanish overnight, for regulatory, geopolitical or legal reasons beyond your reach. And a decision made in Washington can suspend access for European customers within hours.
Which business processes rely on a specific model? Treat every model as a critical supplier, with its own risk level.
Isolate the model layer so you can switch from one provider to another. Portability is your best insurance.
Service-continuity, reversibility and data-retention clauses. Retention varies from one model to the next (here 30 days), a GDPR matter in its own right.
A documented and genuinely exercised plan B: fallback model, portable prompt set, switchover procedure.
Molderez Consult SRL helps Belgian organizations build a resilient, portable AI strategy, so a directive does not turn into an outage.
Assess my AI dependencyTransparence : cet article a été rédigé avec l'aide de l'intelligence artificielle, puis relu par Molderez Consult SRL.