White House Opposes Anthropic’s Plan to Expand Access to Mythos Model

Anthropic recently proposed letting roughly 70 additional companies and organizations use Mythos, which would have brought the total number of entities with access to about 120, people familiar with the matter said. Administration officials told the company they oppose the move because of concerns about security, the people said. Some White House officials also worried that Anthropic wouldn’t have access to enough computing power to serve that many more entities without hampering the government’s ability to use it effectively, one of the people said.

The White House is involved in the rollout of Mythos because of the national-security risks posed by the model. Its ability to find and exploit software vulnerabilities has rattled government agencies and companies in recent weeks. The discussions between Anthropic and the White House are meant to address the risks and have served as an attempt to repair the relationship between Anthropic and the government.

The move by the White House to rebuff the request of Anthropic to let more organizations use Mythos shows that their relationship is still complicated despite efforts from both sides to de-escalate. The administration tried to cut ties with the company because of a spat with the Pentagon over how its AI tools are used by the military, a fight that is working its way through the legal system in two different court cases.

Anthropic initially gave about 50 companies and organizations managing critical infrastructure access and doesn’t have plans for a public rollout. Some government agencies have access, and the Trump administration is working to expand that.

The administration is balancing innovation and security while cooperating with the private sector to make sure AI models are rolled out safely, a White House official said.

Anthropic said last week that it is investigating potentially unauthorized access to Mythos, adding to fears about a potential avalanche of software bugs.

The administration has criticized Anthropic for its ties to organizations that donate to liberal causes such as the calls for strict AI regulations. The Trump administration has also criticized the company because there are several former Biden administration officials on its payroll.

Analysts including Dean Ball, a former AI adviser in the Trump administration who is now a senior fellow at the Foundation for American Innovation think tank, have said it is counterproductive for Anthropic and the administration to be at odds because they need to work together on topics such as cybersecurity.

Cybersecurity experts said Mythos and cutting-edge AI models from OpenAI and Alphabet’s Google are becoming so good at finding and exploiting bugs that they could make it easier for people to launch cyberattacks.

All three companies have taken steps to give security researchers early access to their models so they can be used to find and patch bugs before they are exploited. Still, the sheer volume of bugs that are being discovered and the job of creating, testing and installing all of the patches they will generate have rattled some.

Anthropic initially said it was making a preview version of Mythos available to top companies and organizations on April 7, setting off a scramble around the world to prepare for a potentially wider release. Some White House AI advisers including the venture capitalist David Sacks have speculated that one reason for the limited rollout is that the company doesn’t have access to as much computing power as competitors. The computing power is needed to carry out the tasks of AI model users.

Anthropic struck deals with Amazon.com, Google and Broadcom recently to increase its access to computing power, though it will take time for the projects to go online. AI companies can give priority to the availability of computing power to certain users such as the government. Many are hopeful more-efficient models will help alleviate constraints.

In another example of the company’s bumpy relationship with the White House, an Anthropic researcher named Collin Burns was being onboarded last week to lead a government office that evaluates the best AI models. But the administration abruptly changed course and hired someone else, people familiar with the matter said.

Top White House officials hadn’t been told about the move to hire Burns and didn’t want anyone from a top AI company in a role that needs to work closely with leading companies, one of the people said.

Write to Robert McMillan at robert.mcmillan@wsj.com and Amrith Ramkumar at amrith.ramkumar@wsj.com