Claude Mythos Finds Only One Curl Vulnerability; Experts Divided on What It Really Means

Anthropic's restricted Claude Mythos AI model found only one low-severity vulnerability when testing the widely-used curl tool, despite the company's claims of discovering thousands of zero-days across various codebases. Curl's lead developer Daniel Stenberg says the results suggest Mythos performs no better than previous AI security tools, though some experts argue the limited findings reflect curl's mature and heavily-audited codebase rather than the AI's shortcomings. The debate intensified after Mozilla reported that Mythos helped discover over 270 Firefox vulnerabilities, though all could have been found by elite human researchers.