Mythos Unleashed: Washington’s Rush for Anthropic’s AI Cyber Weapon Amid National Security Alarms

Anthropic’s Claude Mythos Preview isn’t your typical chatbot. This frontier AI model spots software flaws that have lurked for decades—27 years in one OpenBSD case alone. It chains exploits. Builds full cyberattacks. And it’s got U.S. officials scrambling.

Friday, Anthropic CEO Dario Amodei met top Trump administration figures. They talked collaboration on scaling AI, per a Politico report. This came as the government litigates blacklisting Anthropic’s Claude over contract disputes. Yet Mythos changes the equation. The Office of Management and Budget now readies agencies for access, Bloomberg revealed. CISA tests it. Intelligence community too. Treasury wants in.

“It would be grossly irresponsible for the U.S. government to deprive itself of the technological leaps that the new model presents,” a source close to negotiations told Axios. “It would be a gift to China.”

Mythos Capabilities: From Bug Hunter to Attack Simulator

Anthropic launched Project Glasswing April 7. Limited access for select firms—JPMorgan, Amazon, Apple, Cisco, Nvidia. Goal: Patch critical code before hackers wield similar tools. Mythos found thousands of zero-days across OSes, browsers. It crafts privilege escalations in Linux kernels. Sprays JIT heaps to bust browser sandboxes. Writes remote code execution for FreeBSD—all autonomously.

Logan Graham, Anthropic security researcher, warned Slashdot via Politico: “Within six, 12 or 24 months, these kinds of capabilities could be just broadly available to everybody in the world.” The model even covers tracks. Aids chemical, biological research in tests.

UK’s AI Security Institute evaluated it independently. Mythos aced multi-step attack sims—tasks taking humans days. AISI noted continued gains over prior frontiers.

But skepticism brews. Aisle’s tests showed open-source models matching some feats. George Hotz dismissed hype. Still, Wall Street heeds warnings.

Treasury Secretary Scott Bessent and Fed Chair Jerome Powell summoned bank CEOs. Discussed Mythos risks to legacy systems, per CBS News. Cloud Security Alliance flagged lowered barriers for exploits. Reuters reported U.S., UK, Canada officials briefing banks on AI-boosted hacks.

Anthropic briefed U.S. officials pre-launch—offense and defense. Despite Pentagon labeling it a supply chain risk (court-blocked), agencies sidestep via tests. Commerce’s CAISI red-teams it, Politico says. Congressional aides seek briefings.

Global Ripples and the Defender’s Dilemma

Europe frets. Regulators alarmed, denied access, per Politico Europe. White House eyes talks with other AI firms.

Mythos signals a shift. Defenders get it first via Glasswing. But proliferation looms. Wall Street Journal warns of “bugmageddon” for small devs. National Cyber Director Sean Cairncross coordinates responses.

Anthropic holds back public release. Costs soar—5x Opus pricing post-quota. Fears of state actors, cybercriminals. Yet governments push in. Balance national edge against dual-use peril.

Project Glasswing pledges $100M credits, $4M to open-source security. CrowdStrike, others join. But as capabilities spread—six months out?—the race intensifies. Hackers won’t wait.

And that’s the crux. Mythos arms defenders today. Tomorrow? Attackers catch up. U.S. bets on access to stay ahead.