The Trump administration quietly shut down Anthropic’s most powerful AI models for most of the world—then carved out special access for a select few.
Story Snapshot
- The Commerce Department ordered Anthropic to suspend Fable 5 and Mythos 5 globally for foreign nationals, citing national security.
- An alleged “jailbreak” that could help find software vulnerabilities triggered the first known export-control ban on a specific frontier AI model.
- Two weeks later, the Trump administration granted a narrow exemption letting “trusted” companies and agencies use Mythos 5 under safeguards.
How Trump Officials Shut Down Anthropic’s New AI Models
On June 12, 2026, the U.S. Commerce Department issued a binding export-control directive ordering Anthropic to suspend access to its latest models, Fable 5 and Mythos 5, for all foreign nationals. The order relied on national security powers and claimed a jailbreak could bypass safety filters to find software flaws that hostile actors might exploit. Because Anthropic’s systems could not easily separate users by nationality, the company shut down the models globally to follow the directive.
The directive went beyond foreign users living overseas. It treated any foreign national—inside or outside the United States, including Anthropic’s own foreign workers—as covered by the ban. That meant many researchers, security teams, and customers suddenly lost access to tools they used to find and fix cyber threats. The move marked the first known time the Commerce Department used export-control law to target a specific frontier AI model purely on a national security claim.
What the Government Says Went Wrong With Fable 5 and Mythos 5
Government officials told Anthropic they had seen a “coding task jailbreak” that could bypass the models’ safety system and reveal software vulnerabilities in ways that might help foreign militaries or hackers. The concern fits a broader pattern: since 2024, the Commerce Department has used “catch-all” and national security rules more aggressively to control advanced AI models and chips, even when clear technical proof is not shared publicly. Officials argued they had to act fast, even before a public incident, to keep powerful AI from boosting cyber attacks.
Yet the evidentiary record is thin. Anthropic says it was never given a detailed technical report or formal incident description, only verbal briefings about the jailbreak. Independent analysts and Anthropic’s own red team—who had tested the models for thousands of hours with U.S. and U.K. government help—said the exploit was narrow and already mitigated. They argued similar or worse vulnerabilities exist in other cutting-edge systems, including OpenAI’s GPT‑5.5, without export bans.
Anthropic’s Pushback and the Partial Reversal
Anthropic publicly challenged the idea that the jailbreak proved a “catastrophic model failure.” The company said the exploit used minor weaknesses that security teams already knew existed in other models and did not break its guardrails in a general way. Anthropic also highlighted its “defense in depth” approach: strict safety rules, logging and a 30‑day data retention policy to catch abuse, and partnerships with governments and private firms for red teaming. In their view, the directive rested more on a policy fear than on hard technical evidence.
Behind the scenes, Anthropic rushed staff to Washington to negotiate. Cybersecurity leaders and executives sent letters urging the Trump administration to lift the ban, warning it removed some of America’s best cyber defense tools while rivals like China race ahead. On June 26, Commerce Secretary Howard Lutnick sent Anthropic a formal exemption letter allowing Mythos 5 to be released to about 100 “trusted” companies and federal agencies. He wrote that “appropriate safeguards are in place,” effectively admitting the original risk could be managed rather than requiring a total shutdown.
Precedent, Power, and Why Both Left and Right Are Worried
The Anthropic case shows how deeply the federal government now reaches into frontier AI. Export rules already require licenses for advanced model weights trained at huge scale, and allow tight control over who can use powerful chips and data centers. Using those tools to switch off a single named AI model sets a precedent: Washington can decide, overnight, which private technologies the world may use and which must stay locked inside U.S. borders.
Anthropic is brinding back Mythos/Fable access. With cybersecurity uses disabled/limited. They also confirm that GPT-5.5 nd Kimi K2.7 models were able to find the same security vulnerability which caused enacting export controls. The US Government will now get a pre‑release… pic.twitter.com/BTdpBJapyR
— Lukasz Olejnik (@lukOlejnik) July 1, 2026
Critics across the spectrum see familiar problems. Conservatives who distrust the “deep state” worry that unaccountable bureaucrats can throttle innovation and pick winners and losers without transparent proof. Liberals who fear corporate power see a system where big firms and “trusted” insiders get special access while ordinary workers, students, and independent researchers are locked out. Both sides see a pattern: complex decisions made in secret, justified by vague security language, with ordinary citizens bearing the cost.
What Comes Next: Transparency Fights and AI Security Debates
Several open questions now drive the debate. Freedom of Information Act requests could force release of the Commerce directive’s full text and any technical reports that backed the ban. Publishing the Amazon researchers’ jailbreak report and commissioning a truly independent audit of Fable 5 and Mythos 5 could test whether the safeguards really match the government’s claims. A peer‑reviewed study comparing the alleged exploit with vulnerabilities in GPT‑5.5 and other models might show if Anthropic was singled out unfairly.
For now, the Trump administration has partly lifted restrictions but kept tight control over who can use Anthropic’s most powerful tools. That means ordinary Americans, and allies abroad, still do not know exactly what risk justified such sweeping power—or whether politics, bureaucracy, or corporate rivalry played a role. As AI grows more central to the economy and national defense, fights like this will shape not only technology policy, but public trust in a government many already fear serves elites first and citizens last.
Sources:
zerohedge.com, x.com, cnbc.com, fifthrow.com, gtlaw.com, linkedin.com, forbes.com, facebook.com, reuters.com, wired.com, topaithreats.com, labs.cloudsecurityalliance.org, weex.com
