The U.S. government just drew a hard line around frontier AI. Last Friday, the White House ordered Anthropic to stop exporting its most powerful models, Fable and Mythos, to anyone outside the United States, including foreign nationals on American soil, according to TechCrunch AI. Within roughly 90 minutes of being notified, by some accounts, Anthropic pulled the plug on both models. They’ve been dark for a week.
This is the first real test of whether Washington can use export controls to contain frontier AI. And how it ends won’t just shape Anthropic’s access to foreign markets. It could write the rulebook every other AI lab has to build around.
What actually happened
The Commerce Department issued an export control directive citing unspecified national security concerns. TechCrunch AI reports that two events appear to have triggered it:
- The SK Telecom angle. Anthropic gave a South Korean telecom access to Mythos through its limited partner program. U.S. officials grew alarmed after flagging the company, widely reported to be SK Telecom, as one they suspected had ties to China. The company has denied any China connection.
- The Fable jailbreak claim. Amazon CEO Andy Jassy reportedly alerted the administration after Amazon researchers said they found a way around Fable 5’s safeguards. Anthropic disputes the “jailbreak” label, calling it a narrow, already-patched issue rather than a wholesale defeat of the model’s safety.
Some context on why Mythos is the flashpoint. Since launching it in April, Anthropic has marketed Mythos as something close to a doomsday cyber machine, capable of wreaking havoc online if released too widely. Before the ban, only about 150 vetted companies and government organizations had access. The pitch was defensive: help defenders harden their systems before attackers reach the same capability.
Why this matters
What stands out here is that governments have tried this playbook for decades, and the track record is shaky at best. TechCrunch AI walks through the history, and it’s not flattering for the export-control approach.
In the early 1990s, the U.S. treated strong encryption as a weapon. When Phil Zimmermann released Pretty Good Privacy, or PGP, the Customs Service opened a criminal investigation into him for allegedly violating arms export rules. He fought back by publishing PGP’s source code as a printed book, kicking off the “Crypto Wars.” The investigation was eventually dropped. That fight paved the way for the end-to-end encryption now protecting billions of Signal and WhatsApp users.
The spyware era told a similar story. After Western-made surveillance tools turned up on the phones of dissidents, governments expanded the Wassenaar Arrangement to classify hacking software as dual-use. But the treaty has two built-in holes: not every country signs on (Israel, home to major spyware makers, is a notable absence), and enforcement is left to each nation’s discretion. Firms like the sanctioned Intellexa consortium simply relocated to friendlier jurisdictions. There have been wins, like Germany’s FinFisher shutting down in 2022 after a prosecutor’s probe, but they’re the exception.
The pattern is clear. Controls slow distribution. They rarely stop it.
What to watch next
The standoff between Anthropic and the Trump administration hasn’t broken yet. TechCrunch AI notes there’s a reasonable chance the administration lifts the restriction to keep American AI companies competitive abroad.
For practitioners, here’s what to track:
- Precedent. If a directive can take a frontier model offline in 90 minutes, every lab now has to plan for that risk in its deployment and partner strategy.
- Partner vetting. Expect tighter scrutiny of who gets access to high-capability models, especially through partner programs.
- The competitiveness fight. Restricting U.S. labs abroad hands an opening to rivals elsewhere, and that tension will drive how this resolves.
History says containment by decree is hard to pull off. Whether AI proves to be the exception is the open question. You can find the full account at the original TechCrunch AI report.