Why did the White House pull Anthropic's most powerful AI model?
The administration cites national security. Critics see a troubling mix of politics, AI rivalry, and government power.
On Tuesday, Anthropic released a weakened version of Claude Mythus, the powerful AI model whose ability to identify security vulnerabilities sparked fears of a cyber apocalypse and led to access being restricted to a limited number of organizations. Three days later, the model was no longer available. Not because of a glitch, not because hostile actors managed to exploit it for malicious purposes, and not because of any oversight or blunder, but because of an unprecedented order from the White House. The official reason was national security concerns. But against the backdrop of Anthropic's conflict with the Pentagon, and given Donald Trump's well-documented tendency to target perceived opponents, it is possible that this was retaliation against a company that refused to bend to the administration's demands.
Anthropic unveiled Mythus in April, touting its advanced cybersecurity capabilities, particularly its ability to scan software code and identify security vulnerabilities. Such capabilities can help organizations secure their systems. In the hands of hackers, however, they could become a powerful tool for infiltrating critical systems and disrupting, disabling, or stealing information from them.
Because of these risks, the company decided not to release Mythus to the general public. Instead, access was limited to a small group of roughly 50 trusted organizations, including technology giants such as Google, Microsoft, Amazon, Apple, and Nvidia; cybersecurity firms such as Palo Alto Networks; and financial institutions including JPMorgan Chase and the operator of the New York Stock Exchange. In early June, the group was expanded to include 150 additional organizations, including companies outside the United States for the first time.
At the same time, Anthropic began developing a more limited version of Mythus for public use. This version, called Fable, was launched last Tuesday. It included more restricted cyber capabilities, equivalent to those of the Claude Opus 4.8 model already on the market, along with safeguards designed to limit use in high-risk areas such as biology and chemistry, particularly in relation to weapons development. It also included restrictions on the model's use in developing other large language models (LLMs).
Alongside Fable, Anthropic expanded access to Mythus for trusted and authorized partners without most of the restrictions imposed on the public model, except for limitations related to advanced frontier-model development. Unlike Fable, which may refuse certain requests or redirect users to a weaker model, Mythus employed prompt-modification and parameter-adjustment techniques designed to intentionally produce inferior responses in specific areas.
But the developer community had little time to react to Mythus's limitations. The broader public barely had time to evaluate Fable's capabilities, and security researchers had little opportunity to test whether its safeguards could be bypassed. By Friday, both models were effectively unavailable outside Anthropic, and even inaccessible to some company employees.
The reason was an unprecedented White House order.
"The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
“The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
“Access to all other Claude models is not affected.
“We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible," the company announced on X.
Although the order does not, and legally cannot, prohibit American citizens from using the models, Anthropic said it had no practical way to distinguish citizens from non-citizens at scale. As a result, it blocked public access to both models. The company was also forced to deny access to non-U.S. employees, even though some of them likely played significant roles in developing the technology.
The order was issued under the Commerce Department's authority to restrict exports of certain products for national security reasons. One of the most prominent uses of this authority has been the restriction of advanced chip exports to China, a policy that began under the Biden administration.
According to Anthropic, it received the order at 5:21 p.m. Eastern Time on Friday.
"We received the directive from the government today at 5:21pm (ET). The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or “jailbreaking” Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor 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," the company wrote in a statement.
The national-security rationale appears even more questionable given that the Trump administration has simultaneously eased certain restrictions on AI-chip exports to China, including a deal under which Nvidia reportedly agreed to share 15% of related export revenue with the government.
Another possibility is that the order has little to do with national security at all.
Anthropic has been embroiled in a long-running dispute with the Trump administration over the Pentagon's use of its models. In 2025, Claude was the only AI model authorized for use with classified U.S. military information systems. Earlier this year, however, a dispute emerged after Defense Secretary Pete Hegseth demanded that the Pentagon be allowed to use the model for "any lawful purpose."
Anthropic refused, citing concerns that such approval could enable the development and operation of autonomous weapons or facilitate surveillance of American citizens.
In February, after negotiations failed, Trump announced that Anthropic would be removed from federal systems, and Hegseth designated the company a "supply chain risk." The designation, typically reserved for companies linked to hostile foreign powers, effectively required Pentagon contractors to sever ties with Anthropic.
Meanwhile, the Pentagon signed agreements with competitors including OpenAI and Google, whose models met its requirements regarding operational restrictions.
Anthropic's lawsuit against the administration, filed in late March, remains ongoing, although a federal appeals court has allowed the restrictions to remain in place pending a final decision.
Given the limited public evidence supporting the national-security rationale, some observers have begun to wonder whether the order reflects political retaliation. If so, it would represent another example of government power being used to pressure private companies into compliance and punish those that refuse.
In the coming days, the administration's reasoning will likely come into sharper focus. If the decision was primarily motivated by political considerations, and if the courts do not intervene, it could become a warning sign for businesses across the economy. Companies may increasingly find themselves confronting a difficult choice: present a united front against government pressure, or adapt their operations to the demands of a single, highly unpredictable political figure.
