From Banned AI Pariah to White House Ally: Anthropic's Mythos Model Reshapes U.S. Cyber Defenses

A memo from the White House Office of Management and Budget signals a sharp pivot. Gregory Barbaccia, the office's federal chief information officer, notified top officials at the Departments of Defense, Treasury, Commerce, Homeland Security, Justice, and State. The message: safeguards are coming for Anthropic's Mythos model. Agencies could start using it soon. This, despite a ban that branded the company's tech a national security threat just over a month ago. Gizmodo first highlighted the memo's implications, drawing from a Bloomberg review that detailed the email's subject line: "Mythos Model Access."

Barbaccia wrote, "We're working closely with model providers, other industry partners, and the intelligence community to ensure the appropriate guardrails and safeguards are in place before potentially releasing a modified version of the model to agencies." Punchy. Direct. And a far cry from the rhetoric that painted Anthropic as an enemy within. Reuters confirmed the plans on April 16, noting Mythos's role in Anthropic's Project Glasswing -- a controlled rollout for defensive cybersecurity.

Mythos leaked late last month. Anthropic confirmed it. Too powerful, they said. Cybersecurity risks abound. Hackers could abuse it. So no public release. Instead, Claude Opus 4.7 dropped on Thursday -- "not as advanced," especially in cyber ops. Select firms got previews: Nvidia, Microsoft, Google, Apple, JPMorgan Chase. The Bank of England called urgent talks after a demo, per the Financial Times. Crypto players scrambled for access, fearing cracks in encryption, as The Information reported.

Anthropic briefed U.S. officials first. Offensive cyber tools. Defensive ones too. All on the table. Co-founder Jack Clark verified this at the Semafor World Economy Summit, telling attendees the government got early warnings on dangers. Now, collaboration on mitigations. But rewind. Just weeks earlier, the Trump administration slammed the door.

It started with a Pentagon clash. Anthropic refused tweaks to Claude for wartime use -- hours before U.S. strikes on Iran. No mass surveillance. No fully autonomous lethal weapons. CEO Dario Amodei held firm. The response? A historic label: supply chain risk. First for a U.S. firm. Trump ordered agencies to ditch Anthropic tech. Defense Secretary Pete Hegseth directed a phase-out. Contractors shunned Claude. Anthropic sued, claiming retaliation. AP News covered the March 6 designation; courts blocked some blocks.

Yet cracks appeared fast. Commerce's AI Standards center tested Mythos pre-announcement. Two agencies requested access anyway. Three congressional committees sought briefings. Politico broke that story days before Bloomberg's memo scoop. The White House raced ahead. National Cyber Director Sean Cairncross leads efforts to patch vulnerabilities Mythos exposes -- bugs hidden for decades, now found in hours. Wall Street Journal detailed this "Bugmageddon" push, with Treasury's Scott Bessent and Fed Chair Jerome Powell summoning bank CEOs to test the model. WSJ.

Project Glasswing expands. About 50 orgs now probe software flaws. Anthropic's site spells it out: triage automation, patching. Governments must lead, they argue, to counter AI-fueled threats. Allies like the UK eye similar moves. But tensions linger. Pentagon fallout simmers -- contract disputes, lawsuits. Anthropic ditched a safety pledge under pressure, per MSNBC reports from February. Still, civilian agencies win access first.

So why the U-turn? Mythos terrifies. It finds exploits hackers dream of. China looms in cyber shadows. Officials brief: get ahead or get hit. Vance and Bessent grilled tech CEOs pre-release on defenses. Banks test internally. The crypto scramble underscores stakes -- Mythos could break codes. Public stays sidelined. Agencies gear up.

This isn't just tech policy. It's a power realignment. Anthropic, once vilified, now arms the state against itself. Bans bend to necessity. Cyber arms race accelerates. Agencies deploy guarded versions soon. Mythos previews reshape defenses. But risks? Baked in. Guardrails promise control. Reality may differ.

And the Pentagon? Watching. Suits ongoing. Claude embeds persist, ban or not. White House momentum builds. Anthropic's from pariah to partner -- in weeks. National security demands it. Or so the memo says.