Before Anthropic released Mythos to its launch partners last Tuesday, Vice President JD Vance and Treasury Secretary Scott Bessent got every major tech CEO on a phone call and asked them a question that does not have a comfortable answer: What happens when AI models start tipping the balance of cybersecurity in favor of attackers?

The call, first reported by CNBC, included Anthropic's Dario Amodei, OpenAI's Sam Altman, Google's Sundar Pichai, Microsoft's Satya Nadella, xAI's Elon Musk, CrowdStrike's George Kurtz, and Palo Alto Networks' Nikesh Arora. That is not a casual guest list. That is every person who matters in AI security sitting in one (virtual) room with two of the most powerful officials in the country.

The timing matters more than the meeting itself. This call happened before the Mythos release, not after. The Trump administration knew what was coming and wanted answers in advance. According to people familiar with the discussion, officials specifically asked about the security posture of large language models and what the response playbook looks like if models scale in favor of attackers.

The Same Government That Banned Anthropic Still Needs It

Here is the part nobody in Washington wants to talk about: the same administration that designated Anthropic a supply chain risk and told federal agencies to stop using Claude is now pulling Dario Amodei into private calls about national cybersecurity. The same week that a federal appeals court denied Anthropic's request to temporarily block the Pentagon blacklisting, the company was briefing senior officials on Mythos' "full capabilities, including both its offensive and defensive cyber applications."

That is not a contradiction. That is leverage. Anthropic built something so powerful that even the government trying to punish it cannot afford to be out of the loop.

What This Actually Reveals

This call confirms three things we already suspected. First, the US government does not have a coherent AI strategy. It is banning Anthropic from defense contracts with one hand while summoning its CEO for classified cybersecurity briefings with the other. Second, the threat Mythos represents is real enough to get the Vice President and Treasury Secretary personally involved, not just staffers. Third, every AI lab is now operating in a world where their models are simultaneously products, weapons, and diplomatic tools.

Bessent followed up by calling an emergency meeting with the heads of America's biggest banks this week to address Mythos specifically. That is the Treasury Secretary telling JPMorgan, Goldman Sachs, and Bank of America that their entire digital infrastructure might be vulnerable to a model built by a startup that employs fewer people than a midsize accounting firm.

And then Anthropic shipped the model anyway. To Apple, Google, Microsoft, Nvidia, CrowdStrike, and Palo Alto Networks. Because the best defense against a model that can hack anything is making sure the right companies have it first.

Welcome to the new normal: the government asks permission, and the AI lab says "we will take it from here."

First reported by CNBC. Anthropic declined to comment on the meeting but confirmed it has briefed senior US officials on Mythos' capabilities.