Anthropic’s unreleased Claude Mythos AI has reportedly uncovered significant vulnerabilities in Linux and OpenBSD systems that human researchers overlooked for decades, according to sources familiar with the matter. The AI’s findings, still under review, could have major implications for open-source security infrastructure.
The discovered vulnerabilities allegedly exist in core components of both operating systems, with some dating back to the 1990s. “This demonstrates how AI can augment human capabilities in code auditing,” said a cybersecurity analyst who requested anonymity due to the sensitive nature of the findings.
OpenBSD developers have acknowledged receiving preliminary reports but cautioned that full technical details remain confidential until patches are developed. Linux kernel maintainers declined to comment on ongoing security investigations.
The revelations come as major tech companies increasingly deploy AI for static code analysis. Some experts warn this could lead to an arms race between AI-powered security tools and malicious actors using similar technology.