AI Security Startup Depthfirst Outperforms Anthropic, Finds Major Flaws at One-Tenth the Cost

Anthropic launched its Claude Mythos Preview cybersecurity model on April 7, 2026, saying it could identify high-risk vulnerabilities in major operating systems and browsers. NGINX, Linux and Chrome are central to internet infrastructure. If AI can reduce the cost of finding vulnerabilities, it could reshape the defensive model of open-source projects that have long relied on manual review and limited maintenance resources.

Depthfirst said on May 12, 2026, that it had found vulnerabilities in NGINX, Linux and Chrome that Mythos missed. The NGINX flaw had remained undetected for 18 years since 2008, and F5 patched it on May 13. Depthfirst's computing cost was about $1,000, one-tenth of the $10,000 reported by Anthropic. The startup also launched the Open Defense Initiative, offering $5 million in credits.