US agencies quietly test Anthropic's new model despite Trump ban: report

US agencies and congressional officials are still quietly engaging with Anthropic’s new AI model, Mythos, even after President Donald Trump’s administration barred work with the company.

As per a Politico report, the Commerce Department’s Center for AI Standards and Innovation is testing Mythos’ cybersecurity capabilities, while staff on at least three congressional committees have held or requested briefings on the model’s cyber-scanning features.

The development matters as it reveals how quickly a powerful AI tool can become too strategically relevant to dismiss, especially when cybersecurity is involved.

Mythos was announced by Anthropic on April 7, and the company has described it as its most capable model yet for coding and agentic tasks.

The combination has made it both attractive and controversial inside policy circles.

A banned model that officials don't want to ignore

The contradiction is hard to miss, as on paper, the administration has taken a tough line on Anthropic after the Pentagon cut off business with the company.

Pentagon has labeled Anthropic a supply-chain risk after a disagreement over guardrails governing how the military could use its AI systems.

Yet, it seems that parts of the US government are still examining Mythos behind the scenes.

That does not necessarily amount to open bureaucratic rebellion.

It must be noted that the nature and extent of the government’s engagement with Anthropic remain unclear.

But even with that caveat, the picture is striking as a formal ban has not stopped officials from wanting a closer look at a model that may matter for cyber defense and national-security planning.

Also read- Anthropic-Pentagon clash raises key question: who is to blame if AI kills?

Why Mythos matters in cybersecurity

The reason Washington is still interested appears straightforward.

Mythos is not being treated as just another consumer chatbot. It is being assessed for cyber capabilities that could help identify vulnerabilities and expose how attackers might exploit them.

Commerce Department unit is testing the model’s hacking prowess, while congressional staff has sought briefings specifically on its cyber-scanning features.

Anthropic’s own statements help explain that interest.

In its April 7 materials on Project Glasswing, the company said Mythos would be used with selected partners on work such as local vulnerability detection, binary testing, and penetration testing.

Anthropic also said it has been in ongoing discussions with US government officials about the model’s offensive and defensive cyber capabilities.

What this says about Washington’s AI dilemma

The broader lesson is that AI policy is easier to announce than to enforce cleanly.

Governments can impose bans, blacklists, and restrictions, but those measures become harder to sustain when the technology in question appears strategically important.

Washington is trying to balance two impulses at once: setting limits around powerful AI firms and making sure the US does not fall behind on tools that could matter for defense, cyber resilience, and critical infrastructure.