Anthropic says the US directive tied to a narrow jailbreak concern did not justify a full global shutdown of Fable 5 and Mythos 5.
What happened in plain terms
Anthropic launched Fable 5 as a consumer-facing version of its Mythos-class work, then said the US government instructed a global suspension on June 12 after issuing an export-control directive. The rule required stopping access to both Fable 5 and Mythos 5 for foreign nationals, and Anthropic said it could not apply that requirement narrowly at rollout scale.
The company reported receiving the directive at 5:21 p.m. ET and said the letter did not include specific details on the underlying national-security concern. It described the issue as a narrow, non-universal jailbreak, and said the reported behavior did not justify recalling a model used by millions.
Why this is a policy collision, not just a jailbreak incident
Anthropic’s official statement says no single reported bypass has been shown to cause broad harmful impact, and compares the claim to risks present in other public models. Its position is that broad safety engineering still catches most attacks, and that the company’s safeguards reduced the blast radius of high-risk cases.
The legal move was still a hard stop. Under export-control logic, a narrow finding can still trigger total suspension if authorities and providers cannot enforce the restriction without global rollback. The practical result is that safety posture is now shaped by how a model can be controlled across borders, not only by technical severity.
The legal instrument is broad. The alleged exploit may be narrow.
What is still unclear
Several outlets have reported that Anthropic was told to patch or de-deploy and did not act quickly enough. Anthropic has not adopted that full sequence as its own official framing, and continues to call the move a misunderstanding it hopes to reverse. That leaves room for uncertainty around what was shared, what was disputed, and what exact evidence was used to justify the directive.
For users and enterprise teams, the takeaway is straightforward. Frontier model launches now carry a new operational risk profile: a narrow security finding can still produce a whole-product recall. That changes launch planning, internal risk review, and how quickly teams can treat any new capability as settled.