Anthropic proposed adding about 70 companies and institutions to Mythos, bringing the total to around 120 — but the White House rejected the plan on security and compute grounds. Mythos, launched in early April, can detect and exploit vulnerabilities in a wide range of critical software. For now, it's only being tested by companies managing critical infrastructure, with no public release planned.
The White House is worried Anthropic doesn't have enough compute. Opening up to 70 more commercial users could mean the government itself might not get enough compute to run Mythos. White House AI advisor David Sacks has publicly said that Anthropic's limited release itself shows that its compute capacity trails competitors. Anthropic has signed compute procurement deals with Amazon, Google, and Broadcom, but new capacity isn't online yet.
Politics haven't helped. The Trump administration has criticized Anthropic for hiring multiple former Biden officials and for its ties to liberal organizations. One detail shows just how deep the trust gap goes: Anthropic former researcher Collin Burns was slated for a government AI model evaluation role, but senior White House officials replaced him, saying they didn't want someone from an AI company sitting in a position that deals with all the major AI companies.