The fight over Anthropic’s Fable and Mythos models is bigger than one company’s product access. It is a test of whether the U.S. government can use export controls to manage frontier AI in the same way governments have tried to manage powerful cyber tools before.
So far, the story points to a familiar problem. Software can be restricted, licensed, investigated, or banned from certain markets, but history shows that those controls often struggle when the same capabilities can spread, reappear, or move through different jurisdictions.
What happened to Fable and Mythos
Last Friday, the White House cited unspecified national security concerns and ordered Anthropic to restrict exports of its AI models Fable and Mythos. The restriction applied to anyone outside of the United States, as well as foreign nationals inside the country.
Anthropic then pulled access to both models. They have now been unavailable to anyone for a week.
Mythos had already been treated as a highly sensitive system. Since Anthropic launched Mythos in April, the company had marketed it as a powerful cyber model that could cause serious internet harm if distributed too broadly. Before the ban, access was limited to around 150 vetted companies and government organizations.
The stated purpose of that limited access was defensive. Anthropic wanted approved users to help secure software and services before attackers could reach similar capabilities.
The reported triggers behind the ban
Two events reportedly helped push U.S. officials toward the export control directive.
The first involved Anthropic giving a South Korean telecom access to Mythos through its limited partner program. U.S. officials reportedly became concerned after identifying the company as one they suspected had ties to China. The company, widely reported to be SK Telecom, has denied any China connection.
The second involved Amazon CEO Andy Jassy. He reportedly alerted the administration after Amazon researchers, according to him, found a way around Fable 5’s safeguards.
Anthropic disputes the "jailbreak" label. The company described the issue as a narrow, already-patched problem rather than a broad failure of the model’s safety systems.
Even with that disagreement, the government response was direct. The Commerce Department issued an export control directive, and Anthropic had to move quickly to limit access to its products. By some accounts, the company had roughly 90 minutes after being notified.
Why encryption history matters
This is not the first time the U.S. government has tried to limit the spread of software it considered dangerous. One of the clearest examples came in the early to mid-1990s, when computer scientists were building encryption tools to protect data moving across the internet.
One of those tools was Pretty Good Privacy, or PGP. It could encrypt data so that intercepted messages were extremely difficult to unscramble before reaching the intended recipient.
The U.S. government initially treated PGP as a threat. Officials feared it would make intelligence collection harder by preventing agencies from reading emails as they crossed networks.
To stop PGP’s distribution, the U.S. Customs Service opened a criminal investigation into PGP creator Phil Zimmermann for allegedly violating arms export controls. Zimmermann responded by publishing PGP’s source code as a printed book, helping ignite what became known as the "Crypto Wars."
The investigation was later closed. That outcome helped clear the path for end-to-end encryption algorithms now used by billions of Signal and WhatsApp users.
Spyware controls show the same limits
A later attempt to control cyber tools came after researchers in the early 2010s found Western-made spyware being used against dissidents in the Middle East.
In response, several governments agreed to expand the Wassenaar Arrangement, an international treaty covering dual-use software and technologies with both civilian and military applications. The aim was to classify surveillance and hacking software as dual-use, requiring spyware companies to obtain export licenses before selling abroad.
That framework had two major weaknesses. Some countries do not adhere to the agreement, including Israel, which is home to some of the world’s most active spyware makers. The agreement also relies on each country deciding how to apply the rules to companies within its own borders.
Italy became an example of that gap. For a time, the Italian government allowed Hacking Team, then one of the country’s top spyware makers, to export tools around the world despite its record of selling spyware to oppressive governments that used it to hack journalists and human rights activists.
Other European countries have also been lax with spyware makers. Europe has many spyware and hacking tools makers, and despite numerous scandals it has continually failed to curb spyware exports to authoritarian regimes. Critics say a renewed effort across the bloc of 27 member states "does not go far enough."
Some spyware companies also adjusted by moving. Intellexa, a sanctioned consortium of spyware companies, moved operations to countries with lax export controls. Other spyware makers sought to move operations to Saudi Arabia for similar reasons.
There have been some enforcement wins. Germany-based FinFisher shut down in 2022 after a multi-year investigation by German prosecutors into allegations that the company sold spyware to Turkey without an export license. Investigators had previously found FinFisher spyware deployed on the phones of critics of Turkey’s government.
What the Anthropic standoff could decide
The impasse between Anthropic and the Trump administration remains unresolved. One possible outcome is that the administration lifts the restriction to keep American AI companies competitive worldwide.
That would amount to an acknowledgment that AI labs elsewhere, including in China, are likely to reach similar capabilities regardless of what the U.S. restricts.
Another outcome would push American AI companies into a much heavier compliance model. They could need government approval before serving foreign customers, a burden that would affect their bottom line.
The broader lesson from encryption and spyware is not that dangerous cyber technologies should be ignored. It is that export controls have repeatedly delivered uneven results when applied to software. The Mythos ban now puts that same question in front of frontier AI: whether access limits can meaningfully slow misuse, or whether they simply reshape where powerful tools are built, sold, and controlled.