Washington Can Switch Off the AI You Use Every Day. It Just Did.

OPINION — At 5:21 p.m. Eastern on June 12, the most capable commercial AI tools available to the public stopped working. Not because of a bug. Not because their maker decided to pull them. Because the United States government sent a letter.

Three days after Anthropic launched Claude Fable 5 to the public, Commerce Secretary Howard Lutnick ordered the company to suspend access to Fable 5 and its more powerful sibling, Mythos 5, for "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees." Because no provider can reliably screen hundreds of millions of users by nationality in real time, the only way to comply was to switch both models off for everyone, everywhere. As of this writing, they are still dark.

We write this as people who use these tools every day. So do a great many of our readers, and the banks, hospitals, security teams, and engineering shops they work for. When a cabinet secretary can reach into the market and turn off a general-purpose tool that millions depend on, with no public rule, no hearing, and no evidence shown, that is not a technical footnote. It is a precedent, and it stinks.

What actually happened

The order came as a private letter to Anthropic CEO Dario Amodei, issued under the Export Control Reform Act of 2018, the same legal machinery normally used to keep missile parts and advanced chips out of hostile hands. According to the letter as later published by Bloomberg, Anthropic must now obtain an individually validated government license before "exporting" these models to any destination on earth or to any foreign person anywhere, on pain of criminal and civil penalties. Every other Claude model is untouched. Only the two best ones were singled out.

The stated reason is a "jailbreak." Anthropic says the government has provided only verbal evidence of what it calls "a narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws," and that the same capability is "widely available from other models (including OpenAI's GPT-5.5), and is used every day by the defenders who keep systems safe." The company's position is blunt: it disagrees "that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people."

Read that back. A frontier tool used by hundreds of millions of people was forced offline worldwide on the strength of evidence the government would not put in writing.

This takes power away from the people who use the tools

Strip away the national-security vocabulary and look at the mechanics. There was no published regulation. No notice-and-comment. No hearing. No demonstration of harm that the public, or even the company, was allowed to see. Brian Egan, an export-control lawyer at Skadden and a former State Department legal adviser, called the breadth of the order "unprecedented" and noted these letters are kept secret, leaving "industry and the American public in the dark as to the true nature of the threat." He added that "it is also not clear why Anthropic's models were singled out for this treatment, as compared to other U.S. large language model AI companies."

The sweep is staggering. The order does not block a country. It blocks people by nationality, which means a Canadian engineer in California, a British researcher, an Australian or New Zealand defender, all of them allies, all of them cut off. So were Anthropic's own employees who happen to hold the wrong passport. Ordinary users woke up to find a tool they rely on simply gone, with no appeal and no timeline.

And the people whose actual job is to defend networks are not grateful for the protection. More than eighty cybersecurity executives, including leaders from Nvidia, Adobe, Zoom, and Sophos, signed an open "FreeFable" letter to Secretary Lutnick and National Cyber Director Sean Cairncross urging the administration to ease the restrictions, arguing the curbs hurt the defenders, not the attackers. When the security professionals you claim to be protecting are begging you to stop, you have lost the thread.

Follow the incentives

Here is the part that should bother everyone, regardless of what you think of Anthropic. The alarm that set this off did not come from an independent watchdog. According to the Wall Street Journal, Amazon CEO Andy Jassy personally told Treasury Secretary Scott Bessent and other officials that Amazon researchers had used Fable 5 to obtain information that could be useful in cyberattacks.

Now consider who Amazon is in this story. Amazon is Anthropic's largest investor, having put in roughly $13 billion. It is Anthropic's landlord in the cloud, with Anthropic committed to spending well over $100 billion on AWS. And it is, at the same time, a direct competitor, building its own Nova line of frontier models that AWS markets against Anthropic. Anthropic's biggest backer, its critical infrastructure provider, and its rival are the same company, and that company is the one that walked the complaint into the government.

We cannot prove motive, and we will not pretend to. But you do not need a confession to read an incentive structure. When the entity that triggers a regulator's kill switch stands to gain commercially from a competitor's flagship product going dark, the burden of proof should be on the government to show its work in public. Instead, the public got a secret letter. Pair that with a licensing regime that now lets Washington dangle restoration in near-daily negotiations, backed by the threat of criminal penalties, and the arrangement starts to look less like a safety intervention and more like leverage over one of the country's most valuable companies.

The national-security logic eats itself

Set aside the conflict of interest and take the government's stated worry at face value: that a capable model could help hostile actors find software vulnerabilities. The administration's defenders, including White House AI adviser David Sacks, say Anthropic was warned, "refused" to fix the flaw, and that the company is exposing dangerous cyber capability to people who should not have it. Sacks and others have also raised the prospect that the tool's power could be diverted to adversaries like China or Russia. That concern deserves a serious answer.

But the remedy fails on its own terms. By Anthropic's account, validated against the actual demonstration, the same capability is already sitting in competing models that remain fully online. If that is true, then pulling Fable and Mythos does not deny the capability to a single adversary. It denies it only to the American and allied defenders who were using it in the open, while every bad actor keeps a functional substitute a browser tab away. A national-security action that disarms your own side while leaving the other side fully equipped is not security. It is theater with a body count of legitimate users.

Why this is bigger than one company

It would be easy to file this under "AI industry squabble" and move on. Don't. The thing that was established on June 12 is that a single cabinet official can switch off a widely used, general-purpose tool, worldwide, through an unpublished letter, on evidence no one outside the room has seen, after a complaint from the target's own competitor. Today it is two AI models. The mechanism does not care what the tool is tomorrow.

If the government has a real case, let it make that case in the open, with written evidence, a transparent and fair process, and a standard that applies to every company and not just the one a rival pointed at. That is not a high bar. It is the ordinary price of doing something this drastic to tools that millions of people and businesses rely on. Until Washington is willing to pay it, the right posture for users is not trust. It is alarm, and a demand that these models be turned back on.

This editorial reflects the opinion of the Stake & Paper Editorial Board. Underlying facts are drawn from Anthropic's official statement of June 12, 2026, and reporting by the Wall Street Journal, Bloomberg, Fortune, CNN, Time, NBC News, Nextgov, and legal analysis published in Just Security, June 12–17, 2026. The government's position is represented via on-record statements from administration officials. Anthropic provides the AI infrastructure used elsewhere on this site; that relationship is disclosed here in the interest of transparency.