From The Desk Of The CISO: How Will Anthropic's Mythos Change Vulnerability Discovery?
Barracuda Networks Blog, Wednesday, May 6th, 2026
AI models will discover more vulnerabilities than human defenders within five years, shifting security advantage to attackers.
Anthropic's Mythos research demonstrates that advanced AI models can discover software vulnerabilities and generate exploits faster than human researchers, with state-sponsored groups already using these capabilities.
A five-year predictive model across four scenarios shows a sharp surge in published CVEs in year one, followed by sustained discovery. Most critically, the analysis projects that attacker share of CVE discovery will rise from one-in-three today to 55-72% by year five, meaning defenders will increasingly respond to vulnerabilities attackers have already identified and weaponized.
The strategic focus must shift from finding vulnerabilities faster to fixing them faster, with patch velocity, exposure management, and automated remediation becoming the decisive control points.