Pentagon Called Anthropic a Supply-Chain Risk, Then Trump Officials Told Banks to Test Its Mythos Model

Anthropic's most advanced AI model leaked through a CMS misconfiguration weeks before its planned release. The draft blog post, obtained by Fortune on March 26, described Claude Mythos as a "step change in capabilities" specifically tuned for cybersecurity tasks. The irony wasn't lost on anyone: a security-focused AI revealed through poor security practices.

The leak triggered an immediate market reaction. Cybersecurity stocks plunged $14.5 billion in a single day, according to the Times of India, as investors processed what this meant for traditional security companies. CrowdStrike, Palo Alto Networks, and Zscaler all saw significant drops as traders bet AI would disrupt their business models.

Unlike general-purpose models, Mythos represents Anthropic's first attempt at building an AI system optimized specifically for defensive cybersecurity. The company claims it can identify zero-day vulnerabilities, analyze complex attack patterns, and generate patches at speeds that dwarf human security teams.

According to Anthropic's official blog post released Tuesday, the model underwent extensive red-teaming focused on cybersecurity scenarios. The company states Mythos demonstrates "sharply improved reasoning and coding skills" when applied to security contexts, particularly in finding subtle vulnerabilities in large codebases that traditional scanning tools miss.

The model's capabilities reportedly include automated penetration testing, real-time threat hunting across enterprise networks, and generating defensive countermeasures. However, these same capabilities could theoretically be repurposed for offensive attacks if they fell into the wrong hands.

While Anthropic's public messaging still calls Mythos a "research preview," the company has already begun limited deployments inside Fortune 500 companies. Sources familiar with the rollout say Morgan Stanley, Goldman Sachs, and JPMorgan are running controlled tests on internal networks, though none have confirmed this publicly.

The pilot programs focus on threat detection and vulnerability scanning rather than automated patching, reflecting banks' caution about handing an AI system the keys to production infrastructure.

Here's where things get messy. The Department of Defense quietly added Anthropic to its "supply-chain risk" list in late March, citing concerns about foreign investment and data handling, according to two defense officials who spoke to TechCrunch. The designation doesn't ban government use but triggers extra scrutiny for any contracts.

Yet just days later, Treasury Secretary Scott Bessent and Fed Chair Jerome Powell began personally encouraging bank CEOs to test Mythos. The mixed signals have left Wall Street executives confused about which federal guidance to follow. One bank chief described the situation as "being told to speed up and slow down at the same red light."

The tension between Pentagon caution and White House enthusiasm reflects broader debates about how aggressively the US should adopt frontier AI systems. Anthropic's close ties to Google (which invested $300 million in 2022) and Amazon (which committed $4 billion in 2023) add another layer of complexity given both companies' cloud contracts with the defense sector.

Bank executives say they'll proceed with testing but plan to keep Mythos isolated from critical systems until the regulatory picture clarifies. The Fed has promised "additional guidance" within weeks, though officials privately admit they're still figuring out the rules themselves.

Cybersecurity stocks have recovered slightly since the initial $14.5 billion drop, but investors remain skittish. The new Pentagon designation actually boosted shares of traditional security firms as traders bet stricter AI adoption rules might slow disruption. CrowdStrike rebounded 8% on the news, while Palo Alto Networks gained 6%.

Anthropic itself hasn't raised new funding since the leak, though sources close to the company say several sovereign wealth funds have expressed interest. The valuation remains pegged at $18 billion from its last round, but insiders acknowledge that could change quickly depending on how the federal guidance shakes out.