IMPLICATOR
Anthropic said a Commerce Department order arrived at 5:21 p.m. ET on June 12 and required it to suspend access to Fable 5 and Mythos 5 for foreign nationals, including some of its own employees. By midnight, the company had disabled the models globally, saying it couldn’t reliably comply with the order on a selective basis. One administration official told Axios that Anthropic “came to every fork in the road and took the wrong fork.” Another said, “They screwed us.”
Anthropic described the issue as a narrow jailbreak and a flawed government process. Administration officials, according to Axios and Fox Business, viewed it as a test of the company’s seriousness and national-security judgment.
Key Takeaways
- Commerce forced Anthropic to shut Fable 5 and Mythos 5 to comply with foreign-national restrictions.
- Anthropic treated the dispute as a narrow jailbreak; administration sources treated it as a trust test.
- The order gives future AI launches a political checklist beyond technical red-team results.
- European and cyber leaders warned the shutdown could push allies toward alternative AI suppliers.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
A Shutdown Built for Chips, Applied to Software
The Commerce order applied to foreign nationals “whether inside or outside the United States,” Anthropic said. That scope reached customers, allies and some researchers inside the company.
For a hardware export, regulators can stop a shipment or inspect inventory. For a hosted AI model, there is no box to seize and no simple way to sort every user by nationality. R Street Institute’s Mark Dalton wrote that the order’s deemed-export structure made immediate compliance especially difficult for an online service.
Anthropic said that left it with one practical choice: switch the models off for everyone.
Anthropic described Fable 5 as a public version of a more powerful Mythos-class system, while Mythos 5 remained limited to vetted Project Glasswing partners and selected organizations. The company said the models had gone through thousands of hours of red-team testing by U.S. and U.K. AI safety institutes, private testers and its own teams.
Where the Company Lost the Room
Amazon Chief Executive Andy Jassy called Treasury Secretary Scott Bessent on Thursday to raise concerns that Fable and Mythos could be jailbroken, Axios reported. The Commerce order arrived the following day.
Anthropic disputed that account. The company said it had approval to deploy Fable and that officials provided only verbal evidence of a narrow jailbreak. A person close to the company told Fox Business that Anthropic didn’t refuse to fix the issue, but was asked to pull down Fable within 90 minutes without receiving enough detail.
The administration saw it differently. Fox Business quoted a senior official saying Anthropic’s “recklessness” damaged trust. Axios reported that officials believed the company hadn’t learned how to communicate with the administration or account for its ideological differences.
The Safety Argument Collided With the Business Case
Anthropic had invited unusual scrutiny. In April, it described Mythos as too powerful for broad public release. Two months later, it introduced Fable as a safer version suitable for general use.
Get Implicator.ai in your inbox
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
That shift made the company’s public defense harder. Anthropic said the reported vulnerability wasn’t a universal jailbreak and that no tester had found one. It also said perfect jailbreak resistance probably isn’t possible for any AI provider and that the disclosed technique exposed only “previously known, minor vulnerabilities.”
A company that had spent months warning about the cybersecurity implications of Mythos-class systems now said it “disagree[d] that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.”
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
The company also faced questions about business terms. Stratechery noted that Fable required 30-day data retention, including for some enterprise plans that had previously promised zero retention. Separately, the Fable system card said the model could silently limit some frontier-LLM-development requests, affecting about 0.03% of traffic and fewer than 0.1% of organizations. Anthropic later changed that mechanism to a disclosed handoff to Opus 4.8, but the original language still suggested covert model steering to enforce safety and usage policies.
The Precedent for the Next AI Lab
The Financial Times quoted Helen Toner of Georgetown’s Center for Security and Emerging Technology saying that jailbreaks can’t be fully fixed in today’s models. Anthropic said the reported capability was widely available from other providers, including OpenAI’s GPT-5.5. Cybersecurity figures led by Alex Stamos argued that the shutdown hurt defenders because similar capabilities remain available in rival systems.
The European Commission said the models carried both cyberdefense benefits and risks, but warned that contingency measures shouldn’t discriminate against partners. SiliconANGLE quoted European politicians saying the episode strengthened the case for domestic AI providers such as Mistral.
For other U.S. labs, the episode shows that red-team results and deployment approval may not settle a launch if a major cloud partner and investor like Amazon raises an alarm with the administration.
As of Monday, June 15, Anthropic was still trying to restore access. The BBC reported that executives were expected to meet Commerce officials that day and receive more documentation about the issue. Whether Commerce restores foreign access, and on what conditions, is the next decision other AI labs will be watching.
Frequently Asked Questions
Why did Anthropic take Fable and Mythos offline?
Anthropic said a Commerce Department directive barred foreign nationals from accessing Fable 5 and Mythos 5. Because it could not comply selectively, it disabled the models for all users.
What mistake did Anthropic make?
The draft argues Anthropic misread Washington by treating the dispute mainly as a technical jailbreak issue while administration sources treated it as a test of seriousness, trust and national security.
Was the reported jailbreak unique to Anthropic?
Anthropic said the capability was available from other models, including OpenAI’s GPT-5.5. Cyber leaders also argued similar capabilities exist across leading models.
Why does the order matter for future AI launches?
It shows that frontier AI launches may depend on government confidence in the operator, not only on model cards, red-team reports or safety claims.
What happens next?
Anthropic was expected to meet Commerce officials on June 15 and receive more documentation. Any restart conditions will become a signal for other AI labs.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Marcus Schuler
Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai
