US agencies are ignoring Trump's Anthropic ban: Has Claude Mythos sparked a policy clash?

Anthropic has clarified that it wants to ensure responsible deployment by limiting access to select entities and collaborating with government officials on evaluations.

Back in February, the US Department of Defense and Anthropic had a fallout over the usage of its Claude AI in defence applications. At the time, the US marked Anthropic as a supply chain risk, banning the company's products from involvement in defence applications and several federal departments. US federal agencies, however, have reportedly been eager. to test the new Claude Mythos model for its advanced cybersecurity capabilities.

Claude Mythos was unveiled last week as Anthropic's most powerful AI model to date, with capabilities so advanced that it had to be restricted from the public. Anthropic clarified that the advanced capabilities of the AI model made it suitable for cybersecurity strengthening purposes, helping infrastructure companies identify and exploit unknown software flaws in existing platforms.

The model has impressed and alarmed researchers with its advanced hacking and vulnerability-detection prowess.

Now, according to a report from Politico, it is said that staff from at least two large federal agencies have contacted Anthropic expressing interest in using Mythos for cyber defence. The Commerce Department's Center for AI Standards and Innovation (CAISI) is already conducting active evaluations, including "red teaming" exercises to assess the model's capabilities and national security risks.

US government agencies bypassing Trump's restrictions on Anthropic

Experts have warned that Mythos represents a significant leap in AI-driven cybersecurity, capable of finding flaws in major browsers and operating systems far faster than human teams. Anthropic has clarified that it wants to ensure responsible deployment by limiting access to select entities and collaborating with government officials on evaluations.

Former national security officials have expressed concern that political tensions could undermine US cyber defences at a critical time. "It's ironic that the US government tried to ban US government use of Anthropic products -- and then a few weeks later, there's this revolutionary Anthropic product that's very important for cybersecurity," said Charlie Bullock of the Institute for Law and AI.

The Pentagon has "shot itself in the foot by giving the middle finger to the most capable AI provider," said another one of the three aides, as mentioned in the report.

Anthropic has confirmed it is making Mythos available for government testing and evaluation. The company has clarified that other advanced models with similar capabilities are expected within the next two years, which could affect the country's cyber defences in the meantime. However, the pressure is mounting for clearer guidelines on how the US government can safely harness these powerful tools without compromising security or principles.

Will Trump administration lift ban on Anthropic?

In late February, Trump and Defense Secretary Pete Hegseth ordered government agencies to stop using Anthropic's models after CEO Dario Amodei opposed using AI for autonomous lethal weapons or mass domestic surveillance. The Pentagon later designated Anthropic a "supply chain risk," effectively barring its use on Defense Department contracts.

Anthropic has since then sued the government over this designation, resulting in split court rulings. Despite these restrictions, congressional staff from at least three committees have requested briefings on Mythos, and Treasury Department IT officials are also exploring its use for patching network vulnerabilities.

The White House stated it continues to engage with AI companies to secure software vulnerabilities and protect Americans as its topmost priority. However, critics argue the ban has hindered the government's ability to utilise cutting-edge domestic AI technology against rising cyber threats from adversaries like Russia and China - something which Anthropic can only help in today's scenario.

What remains to be seen is whether the Trump administration relaxes the 'supply chain risk' designation in the interests of utilising the most cutting-edge technology for strengthening its cyber defences.