Anthropic has opened direct lines with the Trump administration on its latest AI model, Mythos, even though the Pentagon recently halted all contracts with the company. Co-founder Jack Clark told a Semafor event in Washington that the disagreement centers on how the military would apply Anthropic’s guardrails, not on national security itself. He stressed the importance of keeping government officials informed about frontier AI work and said Anthropic will brief them on Mythos and future releases. Announced on April 7, Mythos outperforms Anthropic’s previous systems at coding and “agentic” tasks—autonomous actions driven by user prompts. Security experts warn those strengths cut both ways: Mythos could spot software vulnerabilities and even suggest exploits. Anthropic argues that engaging regulators and defense agencies now helps preempt misuse. Last month’s Pentagon decision classified Anthropic as a supply-chain risk, blocking its tools for the department and contractors. Anthropic challenged that move in court. A federal appeals panel in Washington declined to put the blacklisting on hold, though a different appeals court had earlier sided with Anthropic in a separate challenge. Details on which agencies beyond the Pentagon are talking with Anthropic remain unclear.