Anthropic's Mythos AI: Bug Finder Public Release Soon?

Anthropic plans to release its Mythos-class AI models to the public once strong safeguards are in place. Currently, these AI models, which find bugs in software code, are only available to select partners through Project Glasswing. Anthropic acknowledges that no organization has yet developed safeguards strong enough to prevent potential misuse or harm from such powerful models. They did not give a timeframe for the public release. Mythos, first introduced in early April, has already scanned over 1,000 open-source projects. It has identified 23,019 flaws, with 6,202 estimated as high or critical severity vulnerabilities. Of these, 1,752 high or critical vulnerabilities were verified by Anthropic, and 90.6% were confirmed as valid flaws. One critical flaw Mythos uncovered affected the wolfSSL cryptography library, used by billions of devices. This flaw, tracked as CVE-2026-5194, could have allowed attackers to forge certificates. It has since been patched. The high volume of findings from Mythos is creating challenges for software maintainers who must patch these vulnerabilities. This development highlights the ongoing tension between powerful AI capabilities and the need for robust security.