On June 12, 2026, Anthropic disabled its two most capable models for every customer in the world. Claude Fable 5 and Mythos 5 had launched three days earlier, on June 9. They went dark not because of an outage or a safety recall the company chose, but because the US government told it to.
The directive cited national security and ordered Anthropic to suspend all access by any foreign national, whether inside or outside the United States. A company cannot check the nationality of every user in real time, so compliance meant one thing. "The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance," Anthropic wrote.
You can own your prompts, your data, and your application. You do not own the model, and last week that distinction stopped being academic.
What actually happened
The government's stated concern was a method of jailbreaking Fable 5. Anthropic said it reviewed a demonstration and found a small number of previously known, minor vulnerabilities. It disagrees with the decision and says it is complying while working to restore access. Every other model, including Claude Opus 4.8, stayed online, according to MarkTechPost.
So this was narrow in scope and, Anthropic hopes, temporary. That is the reassuring read. The other read is that a government reached into a commercial product used worldwide and switched it off with a few days of warning, and the company that built it could only comply.
Why this one is different
Governments have restricted AI before. Export controls on chips, rules on model weights, and access limits by country are all familiar. This is not that. This is the first time a government has ordered a live, commercially available model pulled for every customer on the planet, citing the conduct of a subset of users it could not isolate.
The precedent is the point. A capability that exists one morning can be gone by the afternoon, not because it failed, but because of a policy decision made somewhere upstream of the company that sells it. We wrote about the control questions these models raise in our look at Fable 5 and capability control. The control turned out to sit even higher up than the lab.
The dependency nobody priced in
For anyone building agentic systems, this is a continuity problem dressed as a news story. An agent that books travel, reconciles invoices, or runs a support queue is only as available as the model underneath it. If that model can disappear for regulatory reasons with no notice, the agent inherits the risk.
This is the same dependency we raised in our piece on European banks building on borrowed AI infrastructure. There the worry was relying on a foreign provider. The Fable 5 order shows the provider's own government is a single point of failure too. A frontier model is infrastructure, and infrastructure you cannot control is infrastructure you have to plan around.
The practical response is unglamorous: build so a model can be swapped. Teams that abstracted the model behind their own layer can route to Opus 4.8 or another option and keep running. Teams that wired one specific model deep into a product spent last week rewriting, as we have noted about model lock-in before.
What to watch
Watch whether access is restored quickly and what the terms are, because that sets the template for the next time. Watch whether enterprises start writing model-continuity clauses into contracts the way they write uptime guarantees. And watch whether this pushes serious buyers toward multi-model architectures, not as a performance hedge, but as a political one.
The infrastructure is ahead of the demand, as ever. What changed last week is that the people relying on it learned how quickly it can be taken away.
Sources
If a model your business runs on can be switched off by a government in an afternoon, is it a product you bought or a permission you were lent?
Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.