White House Chief of Staff Susie Wiles is planning to sit down with Anthropic CEO Dario Amodei. The topic? The company's new Mythos model, released on April 7, which has the federal government's full attention thanks to its unnerving cybersecurity capabilities.
An unnamed White House official said the Trump administration is in talks with several advanced AI labs about model and software security. Any new tech that might be used by the feds needs a specialized technical review. Anthropic declined to comment on the meeting.
This is a notable pivot. Just a couple of months ago, Trump and Anthropic were publicly at odds. In February, Trump posted on social media that the government "won't do business with them anymore." The spat stemmed from a contract dispute with the Pentagon—Anthropic wanted guarantees its tech wouldn't be used for fully autonomous weapons or surveilling U.S. citizens, while Defense Secretary Hagerty insisted the Pentagon had the right to decide how to use it. Hagerty then tried to label Anthropic a supply chain risk, a rare move against a domestic company, prompting Anthropic to sue in two federal courts. In March, a federal judge paused Trump's order for agencies to stop using Claude products.
The Mythos Variable
The release of Mythos changes the equation. Anthropic says the model's capabilities are "extraordinary," outperforming human cybersecurity experts at finding and exploiting software vulnerabilities. It's only available to a select group of vetted organizations.
Reactions are split. Some call it a marketing stunt; others believe the hype.
David Sacks, the former White House AI and crypto policy lead and a past critic of Anthropic, is surprisingly in the latter camp this time. "On cybersecurity, I'm inclined to believe them this time. This feels more real," he said on a podcast. His reasoning? As programming models get more powerful, their ability to find vulnerabilities and chain them together into attack paths grows in tandem.
Global Scrutiny
Interest isn't confined to the U.S. The UK's AI Safety Institute assessed that Mythos goes "a step further" than previous versions and can exploit systems with weak security. An EU Commission spokesperson said Anthropic is in talks with the bloc about its AI products, including high-end models not yet released in Europe.
At this week's Semafor World Economy Conference, Anthropic co-founder Jack Clark offered a longer-term view: Mythos isn't a "special model." Similar systems from other companies will arrive within months, he predicted, and within a year to a year and a half, China will release open-source models with comparable capabilities. "The world needs to be ready for more powerful AI systems to arrive." (See: When AI Models Promise Security, They Might Be Creating a Whole New Problem)