The United States government just did something completely unprecedented in the history of artificial intelligence. It weaponized export controls to legally block foreign nationals from accessing specific, highly advanced software models. The target is Anthropic. The casualties are its brand-new, ultra-powerful models, Claude Fable 5 and Mythos 5.
On June 12, 2026, at exactly 5:21 PM Eastern Time, Commerce Secretary Howard Lutnick delivered an export control directive to Anthropic CEO Dario Amodei. The order demanded that Anthropic immediately suspend all access to Fable 5 and Mythos 5 for any foreign national. This means anyone without a US passport, whether they are sitting in New Delhi, London, or even working inside Anthropic’s own San Francisco office, is legally barred from using the system. For another perspective, read: this related article.
Because Anthropic cannot realistically police the nationality of every single user hitting its servers in real time, the company had to pull the emergency brake. It abruptly turned off Fable 5 and Mythos 5 for every single user globally, including Americans. If you try to use them right now, you can't.
This isn't just a compliance hiccup. It represents an entirely new way the US government intends to police AI, shifting from restricting physical chips to outright seizing control of the software itself. Related reporting on this matter has been shared by ZDNet.
The Hacking Machine Washington Feared
To understand why the Commerce Department moved so aggressively, you have to look at what Mythos actually is. When Anthropic teased its underlying Mythos-class architecture earlier this year, it sent shockwaves through federal agencies.
Anthropic admitted that early versions of this architecture possessed terrifying capabilities in locating critical software bugs. It allegedly discovered a severe vulnerability in a major operating system that had remained completely undetected by human engineers for nearly thirty years. The core Mythos engine is, quite frankly, the most effective automated vulnerability-hunting machine ever created.
The security implications are obvious. If a tool can find a 30-year-old zero-day flaw in minutes, it can be used in two ways. White-hat defenders can use it to patch the world's infrastructure, or black-hat state actors can use it to systematically dismantle the global financial system, power grids, and defense networks.
To mitigate this, Anthropic launched Fable 5 this week with what it called conservative safeguards. When a regular user submits a query related to advanced cyber operations, the system is supposed to catch it. In less than 5% of user sessions, Fable 5 automatically routes the query to a less capable, safer model. For a highly vetted group of infrastructure defenders, Anthropic offered the unthrottled Mythos 5 model.
The Trump administration didn't care about the routing safeguards. Officials had been experimenting with the model internally for weeks and were already spooked.
A Lone Jailbreak Sparks a Global Recall
The immediate trigger for the government's panic was a security report, reportedly originating from researchers at Amazon, which detailed a specific bypass technique. The government claims this jailbreak allows users to completely strip away Fable 5’s safety filters, giving anyone access to the raw hacking power of the underlying Mythos engine.
Anthropic’s public response is unusually hostile for a defense contractor ecosystem. The company essentially called the government's move an unscientific overreaction based on a total misunderstanding of how large language models work.
According to Anthropic, the government provided only verbal evidence of a narrow, non-universal jailbreak. The exploit basically involves asking the model to read a specific piece of source code and fix its software flaws. Anthropic reviewed the exploit report and noted that the model merely identified a handful of minor, already known vulnerabilities.
Furthermore, Anthropic pointed out that OpenAI’s GPT-5.5 can do the exact same thing right now without needing any jailbreak at all. Software engineers and security defenders use these exact capabilities every single day to keep corporate networks safe.
"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic stated. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."
The AI lab’s argument is simple. There is no such thing as a perfectly un-jailbreakable model. Every frontier system has narrow exploits. Forcing a total global recall over a minor code-analysis bypass sets a dangerous, unworkable standard for the entire tech sector.
The Internal Collapse at Anthropic
The wording of the export control directive creates a logistical nightmare for Anthropic's internal operations. By banning all foreign nationals, the US government has effectively banned a significant portion of Anthropic's own elite research team from looking at their own creation.
High-profile tech figures born outside the US work deep within Anthropic's ranks. Think of structural core leads like co-founder Chris Olah, prominent AI researcher Andrej Karpathy, and lead alignment philosopher Amanda Askell. Under the letter of this directive, these individuals cannot legally access, test, or patch Fable 5 or Mythos 5 on US soil without putting the company in violation of federal export laws.
This creates a bizarre paradox. The very people who designed the safety alignment protocols for these models are now legally barred from accessing them to fix the jailbreaks the government is worried about. Former White House AI officials have noted that moving forward, tech companies may be forced to demand strict proof of citizenship from employees and users alike just to grant access to top-tier compute clusters.
The Broken Relationship with the Pentagon
This sudden shutdown didn't happen in a vacuum. Anthropic and the Trump administration have been quietly at war all year.
The relationship completely ruptured earlier in 2026 when the Pentagon attempted to integrate Anthropic’s models into military systems. Anthropic fiercely resisted. The company refused to let the US military use Claude for domestic surveillance programs or fully autonomous weapons systems, citing its strict corporate safety charter.
The Pentagon retaliated by designating Anthropic a supply-chain risk, a blacklisting maneuver that threatens to cut off the startup from lucrative federal revenues. Anthropic responded by suing the US government.
Just days before this export ban, President Trump signed an executive order requiring top AI firms to submit advanced models for government cybersecurity testing before public release. While that order claimed it wouldn't create a mandatory licensing regime, Secretary Lutnick’s surprise export control letter proves the administration is perfectly willing to use backdoor regulatory tools to achieve the exact same result.
What Happens to Your Workflow Right Now
If your business relies on Anthropic’s ecosystem, you need to adapt immediately. The sudden disappearance of Fable 5 and Mythos 5 isn't getting resolved in a weekend. Here is what you should do right now to keep your projects moving.
1. Audit Your API Routing
Check your application code immediately. If you have hardcoded calls pointing to Fable 5 or Mythos 5 endpoints, your applications are currently throwing errors. Amazon Web Services (AWS) has already revoked access across all regions at Anthropic's request. Manually downgrade your active endpoints to Opus 4.8 or Sonnet variants. These older models remain fully operational and are entirely unaffected by the government's export mandate.
2. Transition Defensive Cyber Work to OpenAI
If you were using Mythos 5 for code auditing, vulnerability hunting, or patch generation, you can't afford to wait for Anthropic's legal team to settle its feud with the Commerce Department. Migrate those workloads over to OpenAI’s GPT-5.5 pipeline. As Anthropic noted in its defensive brief, GPT-5.5 possesses comparable code-analysis capabilities and currently faces no similar regulatory restrictions.
3. Prepare for Strict Identity Verification
The era of anonymous, frictionless access to frontier AI models is ending. If you operate an enterprise platform that leverages these high-end models, start preparing your infrastructure for user identity verification. If the federal government forces a citizenship-checking protocol on frontier AI access, you will need to know exactly who your users are and where they are geographically located to remain compliant.
