Anthropic has unveiled Claude Mythos Preview, a powerful AI model currently withheld from public release. The company considers the model too dangerous for general use due to its advanced ability to find and exploit software vulnerabilities.
Claude Mythos achieved a 93.9% score on the SWE-bench Verified coding test. This performance outperforms the 80.8% score recorded by the previous top model, Opus 4.6.
Anthropic launched Project Glasswing to manage these capabilities through a controlled cybersecurity initiative. The company provides the model to a partner coalition including Amazon, Apple, Google, Microsoft, and Nvidia. These partners use the tool to proactively identify and repair critical software infrastructure flaws. Anthropic has already used the model to identify thousands of high-severity vulnerabilities.