The Government Just Pulled the Most Powerful AI Ever Released. Here’s What Founders Need to Understand.

On June 9, 2026, Anthropic launched Fable 5. Independent benchmarks from Vals AI rated it immediately as the most capable AI model ever made available to the public. Developers and enterprises that had been waiting for Mythos-class capability outside of Project Glasswing finally had access. It lasted seventy-two hours.

At 5:21pm ET on June 12, Anthropic received a letter from the US government. The letter cited national security authorities and ordered the company to suspend all access to Fable 5 and Mythos 5 for any foreign national, anywhere on earth, including Anthropic’s own foreign national employees. Because Anthropic has no way to filter foreign nationals from domestic users in real time, the only path to compliance was to shut both models off for everyone. By 10pm that night, access was gone worldwide.

This is the first time in history a government has forcibly taken a publicly deployed frontier AI model offline. The implications for every founder building on AI infrastructure are more significant than the news coverage has captured.

What the Government Actually Said and What It Did Not Say

The letter Anthropic received did not provide specific technical details about the national security concern. What it said was that the government believed it had become aware of a method of jailbreaking Fable 5. The described technique, as Anthropic understood it, involved prompting the model to read a codebase and identify software vulnerabilities.

Anthropic reviewed a demonstration of this technique and found it surfaced a small number of previously known, minor vulnerabilities. The same technique, the company noted publicly, works on OpenAI’s GPT-5.5 and other publicly available models. In a statement that was notably direct for a company mid-compliance, Anthropic said it believed the government had made a mistake.

Then on June 14, Semafor reported what the government had not said in its letter. The directive was reportedly motivated by fears that a group linked to China had already accessed Fable 5. Not a theoretical jailbreak. A specific entity, a specific access event, and a specific national security concern that the government chose not to detail in the order it sent to Anthropic.

That reporting changes the frame entirely. This was not a precautionary measure based on a vulnerability report. It was a response to something that had already happened.

The Backstory That Makes This Make Sense

The Fable 5 shutdown did not come out of nowhere. The relationship between Anthropic and the US government had been deteriorating publicly for months before June 12.

In early March 2026, the Department of Defense classified Anthropic as a supply chain risk. That designation has historically been reserved for foreign adversaries. Anthropic called it legally untenable. Negotiations between the two organizations collapsed without resolution.

Then on June 10, one day after Fable 5 launched, Dario Amodei published a major policy essay arguing that governments should have legal authority to block or reverse the release of frontier AI models that fail independent safety testing. He compared it to the FAA grounding unsafe aircraft. He was publicly making the case for exactly the kind of government intervention that, forty-eight hours later, would be used against his own company’s most capable model.

David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, posted on X that Anthropic had refused to fix the jailbreak after a highly credible partner of both Anthropic and the government came forward with it, and that the company had prioritized the continued offering of the consumer model over safety. Anthropic disputes this characterization entirely.

Whatever the full truth of the dispute, what is visible from the outside is a company and a government that had already lost trust in each other before the most capable model either had ever dealt with went public.

The Precedent Every Founder Should Understand

Before June 12, 2026, the risk of building on AI infrastructure included the usual concerns. A model gets deprecated. A pricing structure changes. A provider pivots. These are manageable disruptions that businesses plan for.

June 12 added a new category of risk that no founder’s contingency planning had accounted for. A government can now reach into a private AI company and switch off its most capable models for every user on earth, with no advance notice, no technical explanation, and no timeline for restoration. The order arrived at 5:21pm. Access was gone by 10pm. Businesses that had built workflows on Fable 5 had four hours and forty minutes of warning.

This is not an argument against using frontier AI. It is an argument for understanding the full risk profile of the infrastructure you build on. Every tool in your stack sits on top of decisions being made by companies, governments, and the relationships between them. The more powerful the tool, the more consequential those decisions become.

The founders who take this seriously will build with redundancy in mind. Not because Fable 5 being offline is catastrophic for most workflows—it is not, Opus 4.8 and every other Anthropic model remained fully available—but because the principle that government can intervene in AI deployment at this level is now established. The first time it happens is always the precedent that matters.

What Happens Next

As of June 15, Fable 5 and Mythos 5 remain offline. Anthropic committed to publishing a detailed technical rebuttal of the government’s jailbreak assessment. That rebuttal has not yet appeared publicly. When it does it will be the clearest signal of how quickly the situation might resolve and whether the two organizations can find a path back to functional cooperation.

Anthropic has said it intends to restore Fable 5 as a standard part of subscription plans as quickly as capacity allows once the situation is resolved. There is no timeline. The government has not indicated what resolution looks like from its side.

What is clear is that the most capable AI model ever made publicly available lasted three days. The question of who controls frontier AI—the companies that build it, the governments that regulate it, or the adversaries that attempt to access it—is no longer abstract. It is the central question of the AI era and it just got answered in the most concrete way possible.

Founders who understand that are building differently than the ones who are waiting for things to go back to normal. There is no back to normal. This is the new normal.

---

Also read: The Agent Economy Is Already Here. Here’s What It Looks Like.