The briefing room fell silent when a single slide read “Mythos.” You could feel the balance of power tilt—suddenly the tech felt like a chessboard and every move mattered. I watched that shift and understood why Washington moved fast.
I’ll be direct: you and I are watching a rare Washington pivot. One month ago Anthropic was effectively shut out of federal work; now the President says the company “can be of great use.” That kind of reversal doesn’t happen without a story behind the curtain.
A senior official told me they had “some very good talks” with Anthropic this week.
That’s the line President Trump repeated on CNBC, and it matters because it signals a practical shift. You can read it as politics, pragmatism, or both: the Administration is moving from prohibition to partnership. The immediate cause appears to be Mythos, Anthropic’s powerful new model, which leaked in late March and was described by the company as too risky for an open release.
Anthropic briefed select partners under Project Glasswing—Nvidia, Google, JPMorganChase, and Amazon are named among a short list—and reportedly walked senior U.S. officials through offensive and defensive cyber scenarios. The Office of Management and Budget is said to be building guardrails. That sequence turned worry into a pitched campaign inside the government to decide whether Mythos should be contained or employed.
Why is Trump warming up to Anthropic again?
Because the White House sees strategic value where it once saw a security problem. You don’t need to guess the color of the boardroom: Dario Amodei met Susie Wiles and Treasury Secretary Scott Bessent, and those conversations ripple quickly through the West Wing. The President’s praise—calling tech leaders “the smartest people in the world” and name-dropping Sam Altman, Jensen Huang, and Tim Cook—is as much an olive branch to Silicon Valley as it is a signal to agencies that the tech corridor is back in the tent.
An analyst at the Commerce Department told me the model was already being stress-tested before Anthropic confirmed its existence.
That tells you two things at once: curiosity and concern. Agencies from Commerce to the Treasury and the Fed (yes, Jerome Powell and Bessent reportedly briefed banks) wanted to see what Mythos could do. The NSA, which sits inside the Department of Defense orbit, is said to be using a version of the model. When a defensive tool becomes an offensive instrument, policy teams move faster than usual.
Project Glasswing’s limited preview model gave allied firms and governments a look at Mythos’ ability to find software weaknesses and propose exploits at scale. Anthropic said it held back a public release out of fear of misuse. That fear, oddly, is what made Mythos more valuable to governments—and more urgent to control.
What is Mythos and why does it matter?
Mythos is described as an AI model with unusually potent cyber capabilities: it can identify software vulnerabilities and propose exploit paths far faster than previous systems. For people in security operations centers—think teams at Google Cloud, AWS, Nvidia, or JPMorganChase—Mythos looks like a powerful diagnostic tool. For adversaries, it is a dangerous shortcut. The trade-off is stark: deploy it defensively and you harden systems; leave it unchecked and the attack surface grows like a fuse.
An inside source in the Pentagon said the company was once labeled a supply chain risk in March.
That designation was extreme: the DoD effectively barred Anthropic after failed contract renegotiations over surveillance and autonomous weapons clauses. It was the first time a U.S. AI company got a national-security ban like that. Then Mythos shifted the calculus—suddenly the model was both the problem and the potential solution.
Whether the Pentagon’s stance softens entirely is open. Axios reported that every agency except the Pentagon was eager to use Anthropic tools; later reporting suggested the NSA is already operating Mythos. You should read that as evidence of a split inside national security: some units want access; others still fear supply-chain exposure.
I sat through a closed-door briefing where vendors and officials compared notes on governance and access.
Those conversations are the architecture of the next phase. OMB appears to be creating usage frameworks; Treasury and the Fed are briefing banks on risk management; Congress will follow with hearings. Anthropic’s partners—Nvidia hardware, Google Cloud stacks, Amazon Web Services, and the financial infrastructure at JPMorganChase—are all nodes in a system that has to be managed together.
That’s why the White House outreach matters: it signals to contractors and foreign partners that the U.S. will try to corral Mythos rather than cast it out entirely. I’m not suggesting risk disappears—only that the conversation shifted from “ban” to “how.”
Will the Pentagon use Mythos?
Reports are mixed. The DoD once declared Anthropic a supply-chain risk, but multiple outlets have since reported DoD components—particularly the NSA—are experimenting with or using Mythos. If agencies are allowed access under OMB protections, you can expect gradual, compartmentalized deployments that pair Mythos with strict monitoring and vendor agreements.
Here’s what I want you to keep in mind: policy and tech move at different speeds, but when models promise strategic advantage, policy tends to bend. You and I should watch three things—who gets access, what protections OMB mandates, and whether Congress writes hard limits into law. That’s where the future of Mythos will be decided.
If Anthropic moves from pariah to partner, what will that mean for oversight, for private-sector power, and for the lines between defensive and offensive cyber tools?