On Friday, June 12, at 5:21pm Eastern, the US Commerce Department sent Anthropic a letter ordering it to cut off access to its two most powerful AI models — Fable 5 and Mythos 5 — for every foreign national on earth, including Anthropic's own non-citizen employees. The scope was so broad the company had no practical choice but to switch the models off for everyone, three days after launch. The stated reason traced back to a competitor's claim that it had jailbroken the model. No public evidence has been produced.

The watchout: This is the first time the US government has reached into a commercial AI product used by hundreds of millions of people and turned it off — not for something it did, but for something it might do. If your business runs on a single AI vendor, that precedent is now part of your risk profile, whether you've written it down or not.

If someone on your leadership team needs to see this, forward it their way.

Three days before the letter, on June 9, Anthropic had launched two models. Fable 5 was the public-facing one — its most capable model ever, available to paying customers and enterprise accounts, wrapped in safeguards designed to block high-risk requests in areas like cybersecurity and biology. Mythos 5 was the same underlying model with those safeguards loosened, restricted to a small set of vetted partners. The launch was a milestone: Anthropic was demonstrating, weeks after confidentially filing for an IPO at a $965 billion valuation, that it could ship frontier intelligence at scale without the worst-case misuse scenarios.

That distinction between the two models is the whole story, so it's worth being clear about what a "jailbreak" actually means here. Fable and Mythos are the same brain. The only thing separating them is the layer of guardrails Anthropic built around Fable to keep it from answering dangerous questions. A jailbreak is when someone finds a way to talk past those guardrails — to get the safe, public version to behave like the unrestricted one. So the claim that landed on the government's desk wasn't that someone had broken into Anthropic's systems. It was that someone had figured out how to turn Fable back into Mythos.

Then came the letter. Commerce Secretary Howard Lutnick invoked export-control authority — the same legal regime that governs the sale of weapons and sensitive technology to foreign powers — to bar Anthropic from making Fable 5 and Mythos 5 available to any foreign national, anywhere, including those working inside the United States and inside Anthropic itself. Faced with a directive that broad, the company disabled both models for its entire customer base to ensure compliance. Its other models — Opus, Sonnet, Haiku — stayed online. Only the two newest, most capable ones went dark.

According to an administration official who spoke to Axios, the trigger was another company claiming it had found a way to jailbreak Fable — and the administration had already tried, unsuccessfully, to get Anthropic to delay the launch. The export letter was what it reached for when the company shipped anyway.

Anthropic's response was unusually direct for a company that depends on government goodwill ahead of a public offering. It said the government provided only verbal evidence of a narrow, previously known vulnerability — essentially asking the model to read a codebase and flag software flaws — and that the same capability is freely available from other models, including OpenAI's GPT-5.5. It called the action a misunderstanding, said it was complying while it disputed the finding, and warned that if this standard were applied across the industry, it would halt new model deployments from every major provider.

The pre-IPO market reacted the way you'd expect. Anthropic's synthetic shares on secondary markets fell 3.7% over the weekend. But the stock move isn't the story. The precedent is.

Forget the model names and the IPO for a second. A government agency decided a commercial software product was too dangerous to stay on the market, and it was gone within hours. No hearing. No published evidence. No court looking at the claim first. One letter, on a Friday evening, citing a national security concern that the letter itself never spelled out.

There's a reasonable version of this. Governments should be able to pull genuinely dangerous technology from the market, and frontier AI models can do real damage in the wrong hands. Anthropic itself has argued publicly that the state should have the power to block unsafe deployments. Almost nobody serious disputes that in principle.

The problem is the process. The evidence was verbal. The vulnerability, by Anthropic's account, was narrow and already public. The capability in question exists in competing models that are still online today. And the order showed up right after the company turned down a request to delay the launch — which makes it look less like a careful safety call and more like leverage. When the same agency that regulates you can pull your product over an unproven claim, days after you said no to an informal ask, the gap between safety enforcement and arm-twisting gets very thin.

For a traditional business, the politics are secondary. What matters is the mechanism that just got demonstrated: a tool that hundreds of millions of people could use on Thursday was unavailable by Friday night, by government order, with no warning and no transition period. That mechanism now exists. It has been used once. It will be easier to use the second time.

So here's what this should make everyone who isn't Anthropic stop and think about. If you've built a workflow, a product feature, or a customer-facing service on top of one specific AI model, what's your plan for the morning it isn't there?

Most companies can't answer that. They've come to treat their AI vendor like the electricity supply — always on, never really thought about. The Fable shutdown is a reminder that it isn't the electricity supply. It's a commercial product from a private company, operating in a regulatory environment that's being made up in real time. Models get pulled. Terms change. Access disappears over fights you aren't part of and can't influence.

The companies that will weather this aren't the ones who picked the "right" AI vendor. They're the ones who built so they could switch vendors without rebuilding everything.

In practice, that means a few specific things. Don't hard-wire your operations to a single model's quirks; keep your prompts and integrations portable enough to point at a different provider. Know your fallback before you need it — if your primary model vanished today, which one would you move to, and how long would the switch take? Keep a human-legible record of what your AI tools actually do in your business, so that replacing one isn't an archaeology project. And for anything mission-critical, treat single-vendor dependency the way you'd treat a single supplier for a part you can't build without — as a risk to be managed, not a convenience to be enjoyed.

None of this is exotic. It's the same supply-chain thinking traditional businesses have applied for decades to every other critical input. AI just got added to that list, and most companies haven't updated the list yet.

Anthropic will probably get its models back online — it has the firepower to sort this out. Most businesses downstream don't have that luxury. If you're building key business processes around a heavy reliance on one model provider, you need to start thinking about redundancies.

The AI you depend on answers to forces that have nothing to do with you — competitors, regulators, politics, the timing of an IPO. Build like the model could go dark on a Friday, because one just did.

Know someone making AI decisions who should be reading this? Forward it their way.

- Hashi