Anthropic appears to be planning a wider release of its AI security agent, Claude Mythos, after all. Initial signs of a rollout for enterprise customers have appeared in Claude Code and Claude Security.
Earlier this month, Anthropic launched Claude Mythos, the much-discussed AI security agent. The model was supposedly only available to a handful of companies to ensure safety. Now, signs of a broader rollout are emerging, according to The Register. Specifically, Claude Mythos is being prepared for a rollout to enterprise customers in Claude Code and Claude Security.
Risks
Last month, a disagreement arose between Anthropic and the White House. The AI company wanted to make its AI security agent more widely available to multiple parties, but the White House did not agree.
The model caused quite a stir in the security world, with fears that it could also be misused. During the launch, Logan Graham, Frontier Red Team Lead at Anthropic, spoke about the risks of the model: “Such an AI model can also cause harm if it falls into the wrong hands. That is why the model is not available to everyone.”
Claude Code and Claude Security
Meanwhile, signs of a broader rollout are surfacing. Specifically, a recent update from Anthropic states that the company will consider “releasing Mythos-class models in the future.”
TestingCatalog noticed the Mythos 1 model label in the Claude UI before it disappeared. The model is reportedly being prepared for a rollout in Claude Code and Claude Security.
10,000 vulnerabilities
In the update, Anthropic also mentions the current state of the model and how it has helped partners detect vulnerabilities. “After one month, most partners have each found hundreds of critical or severe vulnerabilities in their software. Collectively, they have found more than ten thousand,” Anthropic stated in its announcement.
For instance, Microsoft recently deployed Claude Mythos to detect vulnerabilities, and Mozilla also started working with Mythos Preview. “We and our approximately 50 partners have since used Claude Mythos Preview to find more than 10,000 high or critical severity vulnerabilities in the world’s most systemically important software,” according to Anthropic.