Anthropic shuts down top AI models after US restricts foreign access

Anthropic had to "abruptly disable" Fable 5 and Mythos 5 after the US issued an export control directive ordering the company to suspend all foreign access to the models.

The directive orders Anthropic to suspend access to Fable 5 and Mythos 5 to any foreign national, whether inside or outside the US, including the company’s employees. As a result, Anthropic had to disable access to the models for all its clients.

The company said that its other models will not be affected.

Anthropic explained that it had been given only "verbal evidence of a potential narrow, non-universal jailbreak". It believes the US government became aware of a method of bypassing, or “jailbreaking” Fable 5 to identify software vulnerabilities.

“These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass,” Anthropic said.

A report from The Wall Street Journal revealed that the government’s decision was prompted by conversations between Amazon CEO Andy Jassy and US officials, including Treasury Secretary Scott Bessent. According to the publication, Amazon researchers used a series of prompts to get Fable 5 to provide information that could aid cyberattacks.

In the announcement, Anthropic noted that it had instituted strong safeguards that greatly reduce the possibility of Fable 5 being misused that way. It claimed that the tests conducted with internal organizations, teams, and the US government showed that Fable’s safeguards are substantially more effective than those of any previously deployed model.

No perfect resistance

While no company can guarantee perfect jailbreak resistance, no tester has yet been able to identify a universal jailbreak capable of broadly bypassing Fable 5’s security safeguards, Anthropic said.

Assuming perfect resistance is not possible, the company had adopted a defense-in-depth strategy, aiming to make jailbreaks either narrow or very expensive to develop. Anthropic cited this as the reason for requiring 30-day retention of customer data with Fable – allowing it to monitor for successful attacks, as well as research and mitigate jailbreaks.

“We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result. The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift,” Anthropic added.

The company said that it had reviewed a report that it thinks triggered the government’s directive. According to Anthropic’s investigation, this level of capability is available from other models like OpenAI’s GPT-5.5 and is used by defenders for everyday security tasks.

According to Anthropic, the move does not adhere to the government’s core principles for blocking unsafe AI deployments, and they believe the decision stems from "a misunderstanding."

“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 wrote, adding that applying the rule across the industry would eventually halt all new model deployments.

Political fallout

The directive comes amid heightened tensions between Anthropic and the Trump administration. Earlier this year, Trump ordered the US government to immediately stop using Anthropic’s technology and declared the company a “supply chain risk” over Anthropic’s refusal to loosen ethical and safety guidelines on its Claude AI.

The company ended up suing the Pentagon for placing it on a national security blacklist and has secured at least one early round win in the ongoing trial.

Despite the restrictions, some government agencies have found a way to continue accessing Anthropic’s models. The US military also used Anthropic’s AI tools during strikes on Iran within hours of Trump’s ban.

Regardless, the directive could potentially be a major blow to the company’s plans to go public this year. In June, Anthropic confidentially submitted a draft registration statement to the SEC, taking the first formal step toward a potential IPO.

---

Unlock more exclusive Cybernews content on YouTube.