Anthropic reports that its Mythos model has identified tens of thousands of potential vulnerabilities across widely used software and multiple projects.

Test results and disclosure status

In one controlled test, Mythos reportedly discovered 271 issues in Firefox during a single automated scan, Anthropic said.

According to the company, the majority of the findings have not been publicly disclosed yet, and a significant portion remains unpatched by maintainers.

Model capabilities

Anthropic states that Mythos is already capable of both detecting weaknesses in code and autonomously simulating cyberattacks to verify exploitability.

The company describes the approach as an automated auditing process intended to surface issues at scale and prioritize reviews for security teams.

Timeline for widespread use

Anthropic estimates that tools with offensive simulation capabilities could reach broader availability within 6–12 months, the company said.

This projected timeline, combined with current disclosure and remediation lags, may extend the window during which vulnerabilities can be exploited before fixes are applied.