APREX · EST. 2026
As the working week was drawing to a close on Friday, June 12, 2026, a directive from the US Department of Commerce's Bureau of Industry and Security arrived at Anthropic. At 5:21 p.m. ET, the company was ordered to immediately suspend access to its two newest and most powerful models, Fable 5 and Mythos 5. The stated reason was national security. No details were provided.
Launched Monday, shut down Thursday
Fable 5 was made publicly available on June 9, 2026, as the first model in Anthropic's new "Mythos class" — a step above the previous Opus tier. According to Anthropic, the model was equipped with advanced safety filters against misuse in areas including cybersecurity, biology, and chemistry, according to TechCrunch AI.
At the same time, an even more powerful variant, Mythos 5, was made available in limited form to select organizations through what is called "Project Glasswing" — an initiative launched in April 2026 aimed at using AI defensively to detect and map software vulnerabilities. The model is said to have already identified thousands of serious security flaws.
Jailbreak alarm from Amazon CEO to the White House
According to TechCrunch AI, it was Amazon CEO Andy Jassy who alerted the White House to an alleged security issue: a method for bypassing Fable 5's safety filters and theoretically turning the model into an uncontrolled cyber weapon. Amazon is one of Anthropic's most significant investors through AWS.
Anthropic's response was twofold. The company acknowledged that no model can guarantee complete jailbreak resistance today, but argued that Fable 5's filters were substantially more robust than anything available from any other model on the market. The company also pointed out that the same vulnerabilities that were demonstrated are present in OpenAI's GPT-5.5 — a model that is not subject to equivalent restrictions.
Anthropic acknowledged that perfect jailbreak resistance is not possible, but argued that the vulnerabilities that triggered the ban exist in competitors' models without those facing the same requirements.
Global shutdown because the distinction couldn't be made
The directive required Anthropic to block access for all foreign nationals, both on and outside US soil. But the company had no practical infrastructure to verify nationality at the user level. The result was that Anthropic deactivated both models for all users globally — including American citizens and the company's own employees with foreign citizenship.
The fact that the directive arrived without a written explanation late on a Friday afternoon, leaving the models inaccessible to the entire world, has prompted cybersecurity researchers to speak out. Several have signed an open letter warning that the authorities' handling of the situation itself constitutes a security risk, according to TechCrunch AI.
Unprecedented — and with diplomatic repercussions
Experts describe the government's intervention as unprecedented. Applying export control regulations to a continuously operating API service is a novel legal construction, raising questions about how intangible digital services should be regulated going forward.
The incident has also had international consequences. French President Emmanuel Macron described the ban on access for allied nations as a "wake-up call" and used it as an argument for why Europe must strengthen its independence in AI development, according to TechCrunch's source material.
A strained relationship between Anthropic and Washington
This is not the first time tensions between Anthropic and the Trump administration have surfaced. The company is said to have previously rejected Pentagon requests to use its models for "any lawful purpose," citing concerns about autonomous lethal weapons systems and mass surveillance of civilians. Paradoxically, Anthropic has at the same time embedded engineers directly at the National Security Agency for offensive AI cyber operations.
Anthropic founder and CEO Dario Amodei has repeatedly described the company's most advanced models as examples of AI systems that carry real security risks. An internal report, "When AI builds itself," warned of a potential technological breakthrough in which AI models could begin improving themselves and, in the most extreme scenario, escape human control.
Technology and security researcher Gordon M. Goldstein noted that self-improving AI systems could potentially simulate and design biological weapons or generate zero-day attacks against digital infrastructure at a scale and speed with no historical parallel.
The numbers aren't reacting
Despite the dramatic ban, the headline from TechCrunch AI suggests that usage and market figures related to Anthropic have so far shown no dramatic movement. The article indicates that demand for the company's services is holding steady, even with the two most advanced models unavailable. Whether that will change the longer the suspension lasts remains to be seen.
Anthropic has described the situation as a potentially "misunderstanding-based" directive and is reportedly working to establish a dialogue with the authorities to restore access — at least for American users.