Anthropic announced that, following a U.S. government directive citing national security reasons, it must suspend access to its models Fable 5 and Mythos 5 for anyone considered a “foreign national.” The order arrived on June 12, 2026 and, as an immediate effect, the company disabled those models for all its customers. The rest of Anthropic’s models are not affected.
What happened and how they presented it
The company received the notice the same day at 5:21 pm (ET). The government letter did not include technical details; according to Anthropic’s public communication, the authority believes it identified a method to 'jailbreak' Fable 5.
Anthropic reviewed a demonstration and says the finding shows small, already-known vulnerabilities that other public models can also reproduce without a specific bypass. In short: they found a narrow case, not a universal flaw that wipes out all safeguards.
Anthropic says it is complying with the directive, but disagrees with the measure as disproportionate to what was revealed.
What Anthropic says about safeguards
Anthropic repeats several key points about Fable 5:
- They implemented strong safeguards that, the company says, greatly reduce the risk of misuse related to cybersecurity.
- Before launch they ran red-teaming with U.S. agencies, the U.K. intelligence service and private third parties for weeks and thousands of hours.
- Their tests indicate Fable’s defenses are more effective than those of models deployed earlier.
- So far no universal 'jailbreak' has been found that broadly circumvents the safeguards.
- Anthropic assumes perfect resistance to jailbreaks isn’t attainable today; that’s why they apply defense in depth: limit the breadth of attacks, make large-scale attempts costly, and maintain strict monitoring.
- Because of that monitoring they required 30-day customer data retention in Fable to be able to investigate and mitigate incidents.
They also said what they communicated to the government corresponds to a narrow case that basically asked the model to read code and fix bugs. Anthropic claims similar capabilities exist in other models on the market, and that they will share more details within the next 24 hours.
What this means for users and for the industry
For you as a user or a company: Fable 5 and Mythos 5 are disabled while Anthropic complies with the order. Other Claude products remain operational.
For the industry: the action sets a relevant precedent. Anthropic warns that if the criterion to pull models is merely identifying a narrow jailbreak, the rollout of new architectures across frontier providers could be stifled. Could this be the start of stricter, more opaque control over AI deployments? That’s the question many will ask.
The company also makes a public claim: the government should have the power to block unsafe deployments, yes—but within a legal, transparent process based on verifiable technical facts.
Immediate concrete implications
- Access: disabled for Fable 5 and Mythos 5 for customers, including Anthropic’s foreign employees.
- Transparency: Anthropic says it will share more data in 24 hours; the government has not published the written evidence in the public notice.
- Regulatory debate: pressure grows to define clear review and blocking processes for models on national security grounds.
The story is changing fast. Anthropic apologizes for the interruption and is working to restore the service as soon as possible, while asking for a more transparent and technical process to avoid blanket decisions based on narrow findings.