The US government is quietly preparing to roll out Anthropic’s powerful frontier AI model, Mythos, across federal agencies even though the Trump administration previously cut ties with the company.
Anthropic unveiled Mythos earlier this month under Project Glasswing, a controlled program granting select organizations access to the unreleased Claude Mythos Preview model. The AI has already proven its muscle, reportedly uncovering thousands of vulnerabilities in operating systems, browsers, and widely used software. Its ability to write and analyse code at a high level makes it both a defensive asset and a potential risk if misused.
Gregory Barbaccia, federal chief information officer at the White House Office of Management and Budget (OMB), confirmed in an internal email that safeguards are being set up to allow agencies to experiment with Mythos. He emphasised that the administration is working closely with model providers, industry partners, and the intelligence community to ensure “appropriate guardrails” before releasing a modified version of the tool.
Officials have stressed that any new AI technology must undergo rigorous evaluation for fidelity and security before being integrated into government systems.
Anthropic’s relationship with the federal government has been rocky. Co-founder Jack Clark revealed that the company had discussions with the Trump administration about Mythos, even after the Pentagon severed business ties following a contract dispute. That earlier rupture effectively banned Anthropic from Pentagon work, but the current push to deploy Mythos suggests agencies are willing to sidestep Trump’s ban in favor of addressing urgent cybersecurity needs.
The decision highlights a broader tension: balancing political decisions of the past with the pressing need to secure America’s digital infrastructure today. Mythos’s ability to expose weaknesses in critical systems makes it a double-edged sword capable of strengthening defenses but also dangerous if exploited.
With Reuters inputs
