The Trump administration banned Anthropic from government contracts. Federal agencies said they understood. Then they quietly picked up the phone and called Anthropic anyway.

According to a bombshell Politico report published Tuesday, staff from at least two large federal agencies have reached out to Anthropic in recent days to express interest in integrating Claude Mythos into their cyber defense operations. The Commerce Department's Center for AI Standards and Innovation (CAISI) is already actively testing the model's hacking capabilities, according to four people with direct knowledge of the discussions.

Let that sink in. The same government that blacklisted Anthropic from federal contracts is now quietly circumventing its own ban because the model is simply too powerful to ignore.

The Ban That Could Not Hold

This has been building for weeks. When Anthropic launched Project Glasswing and started distributing Mythos to roughly 40 vetted organizations, it found thousands of zero-day vulnerabilities in major operating systems, web browsers, and critical infrastructure software. Apple, Google, JPMorgan, and other major institutions immediately signed up.

The problem: the Pentagon had already blacklisted Anthropic over the company's refusal to build offensive AI weapons systems. Trump's administration extended that ban across federal agencies. Anthropic lost its Pentagon appeal. On paper, the government was done with Anthropic.

In practice, career officials took one look at what Mythos could do and started making calls. Politico's sources include a current cybersecurity official, a former Trump administration official, and a former senior national security official. These are not junior staffers going rogue. These are senior government officials deciding that America's cyber defense matters more than a political grudge.

Why This Matters More Than You Think

This is not just an embarrassing contradiction for the White House. It is a signal that the Anthropic ban was never about national security in the first place. If Mythos were genuinely dangerous without guardrails, the government would be trying to regulate it, not use it. Instead, the agencies responsible for actually defending American networks looked at the ban, looked at Mythos, and chose the tool that keeps the country safe.

OpenAI scrambled to release its own GPT-5.4-Cyber model last week, positioning it as the government-friendly alternative. But Reuters reported that even OpenAI acknowledged the release was a direct response to Mythos. The timing says everything: OpenAI built a cyber model because Anthropic built a better one first.

The Bigger Picture

Anthropic is now in the bizarre position of being banned by the government that needs it most. Investors are lining up with offers valuing the company at $800 billion. Its revenue tripled to $30 billion. And the very agencies that are supposed to enforce the ban are instead circumventing it.

The Trump administration now faces a choice that no amount of political theater can avoid: lift the ban and admit the whole thing was performative, or keep it in place and watch America's own cybersecurity agencies fall behind because they cannot use the best tool available. Either way, Anthropic wins. The ban is dead. The government just has not said it out loud yet.