On June 9, 2026, Anthropic released Claude Fable 5 and Claude Mythos 5 — described as the most capable models the company had ever shipped, with advanced reasoning and code analysis capabilities that reviewers had been calling a step-change from anything available before.
On June 12, the US Commerce Secretary sent Anthropic’s CEO a directive to suspend all access to both models worldwide under export control authority. By June 13, both models were offline globally. As of late June 2026, they remain offline. No restoration date has been confirmed.
What happened
The directive instructed Anthropic to prohibit access “by any foreign national, whether inside or outside the United States” — language that, on its face, would apply to a substantial proportion of Anthropic’s own user base and, reportedly, some Anthropic employees. Because Anthropic’s platform does not maintain real-time nationality verification for users, the company could not selectively block foreign national access without taking the models down entirely. So it did.
The reported trigger was a jailbreak — a method of bypassing the model’s safety constraints — though the details of why this rose to the level of a government export control directive, rather than a standard patch-and-redeploy cycle, have not been fully disclosed publicly.
Why this matters beyond the headlines
The Fable 5 suspension is the first time a US government export control directive has taken a commercially deployed AI model offline globally, including for US-based users. It’s a novel category of risk that no development team or technology buyer had meaningfully priced in before June 2026.
The practical consequence — Anthropic’s most capable model becoming unavailable with less than a week’s notice, indefinitely — illustrates something that’s been easy to overlook when AI tools are being discussed primarily as productivity accelerators: they are also a dependency. Depending on how deeply AI is integrated into a development workflow, unexpected availability changes can affect in-flight projects.
What it means for commissioning development
Three things follow from this.
First, development teams with serious AI tooling depth don’t rely on a single model or provider. The teams most affected by the Fable 5 suspension were those that had built workflows specifically around it and had to scramble to reconfigure around Claude 3.7 Sonnet or GPT-4o in hours. Teams with model-agnostic setups adapted without disruption.
Second, the event makes the governance dimension of AI tool selection more concrete. Export controls on AI models are now a real category of operational risk, not a theoretical one. It won’t be the last time.
Third, for buyers commissioning software, it’s reasonable to ask development partners what happens to in-flight projects if their primary AI tooling becomes unavailable unexpectedly. It’s the same question you’d ask about any critical third-party dependency.
More on how we approach AI tooling and resilience in development projects on our AI-assisted development page.