by Open Frontier Staff
When news broke on June 12 that the US government had ordered Anthropic to disable Fable 5 and Mythos 5, a predictable thing happened: the national security hawks reached for familiar language. “Foreign nationals.” “Export controls.” “Critical infrastructure.” The framing was designed to invoke seriousness, to make scrutiny feel unpatriotic.
What the government claimed
Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei on June 12 at 5:21 PM ET, ordering the immediate suspension of Fable 5 and Mythos 5 for all foreign nationals worldwide, including Anthropic’s own employees. The stated concern: that someone had discovered a method of jailbreaking Fable 5, potentially allowing adversaries in China, Russia, or other “countries of concern” to exploit its cybersecurity capabilities.
“Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage.” — Anthropic’s own launch statement, June 9, now being used against it
The government’s argument: Fable 5 is so capable at finding software vulnerabilities that a jailbreak of its safety systems would hand a serious cyberweapon to foreign military intelligence. Fable 5 had been benchmarked as the most capable publicly available model on release day.
Anthropic’s response was swift and pointed. The company said it had received only verbal notice of a “potential narrow, non-universal jailbreak”: not a specific exploit, not documented harm. Anthropic’s own review found the jailbreak amounts to prompting the model to read a codebase and identify flaws, a capability already present in OpenAI’s GPT-5.5 and other publicly available models. The company added that its strongest safeguards run through independent classifier systems that operate separately from the model itself, meaning a jailbreak of the model doesn’t automatically bypass the underlying protections.
It’s retribution, not regulation
To understand why, you need the full timeline.
Feb 2026: Pentagon demands Anthropic allow Claude for “all lawful purposes,” including lethal autonomous weapons and mass domestic surveillance. Anthropic refuses. Contract collapses.
Mar 9, 2026: Trump administration designates Anthropic a “supply chain risk to national security,” a label historically reserved for foreign adversaries. Defense contractors must certify they won’t use Claude.
Mar 2026:Anthropic files two lawsuits alleging unlawful retaliation. A federal judge temporarily blocks the blacklisting, finding Anthropic is likely to succeed in court.
Jun 9, 2026: Anthropic launches Fable 5, the first public release of its Mythos-class model family, with cybersecurity guardrails.
Jun 12, 2026: Commerce Department issues export control directive at 5:21 PM. Fable 5 and Mythos 5 go dark for all users worldwide. Pete Hegseth publicly celebrates Anthropic’s removal from DoD facilities the same day David Sacks urges resolution.
Three days after its biggest model launch. During active litigation against the same government. Using a different agency (Commerce, not Defense) to route around the court injunction blocking the first action. The community’s read: this is a coordinated squeeze, not a good-faith security review.
Defense Secretary Pete Hegseth’s public celebration of Anthropic’s ouster, on the same day AI czar David Sacks was urging a quick resolution, suggests the administration isn’t speaking with one voice. These are not the words of a government narrowly focused on a single jailbreak.
This is political theater. The security framing is the costume.
The Next Generation Could Get Built Somewhere Else
Restricting Fable 5 doesn’t delete the underlying knowledge. It doesn’t erase the training techniques. It doesn’t un-publish the research papers. It just means the next generation of capable models gets built somewhere else, by someone else, with fewer safety commitments. Over 80 cybersecurity executives, including leaders from Nvidia and Adobe, signed an open letter to Lutnick and National Cyber Director Sean Cairncross making exactly this argument, urging the restrictions be lifted.
The real worst case isn’t stagnation. It’s gatekeeping.
The most pessimistic take in the community isn’t that AI progress will stop. It’s that access will become stratified in ugly ways.
The export control directive as written targets “foreign nationals, wherever located,” which in practice means the only way to enforce it on short notice is a blanket global shutdown. That’s exactly what happened. There is no reliable way to segment foreign nationals from US persons in real time across a user base in the hundreds of millions. The global kill switch got pulled not because the government wanted to harm US users, but because enforcement of the stated order made it unavoidable.
The fear going forward: mandatory identity verification for US citizens to access frontier models. International users locked out indefinitely. A two-tier internet where the most capable AI is available to credentialed researchers, government contractors, and enterprise customers, and not to the student, the developer abroad, or the small business owner.
Most observers think competition from non-US models will prevent this from becoming permanent. But the precedent set here, that a frontier model can be deployed to hundreds of millions of users and pulled by government order days later, during active litigation, on verbal evidence of a minor jailbreak, is a different kind of problem entirely.
What this moment actually reveals
There’s a grim irony running through this whole affair. Anthropic spent months warning the world that its models were powerful and potentially dangerous, calling for government oversight as part of a “statutory process that is transparent, fair, clear, and grounded in technical facts.” The government listened to the first half of that sentence and ignored the second.
What we have instead is a case study in using the apparatus of national security to settle political scores, dressed up in language serious enough to make journalists write concerned paragraphs. The people who build and use these systems every day aren’t buying it. They’ve done the analysis. They know what real security concerns look like, and they know what retribution looks like.