AI Found 2,000 Vulnerabilities in 7 Weeks. We've Patched Almost None of Them
CIO, Tuesday, June 16th, 2026
An AI model found thousands of unknown vulnerabilities faster than organizations can patch, upending defense models.
Anthropic's frontier model Mythos identified over 2,000 previously unknown vulnerabilities across major operating systems in just seven weeks, autonomously developing working exploits.
The article warns "the window is gone" for traditional vulnerability disclosure where companies had time to patch before attacks. Over 99% of discovered flaws remain unpatched, creating a gap where attack velocity outpaces remediation.
Project Glasswing gives early access to elite firms like Microsoft and JPMorgan, but this two-tier approach leaves mid-market enterprises exposed. Organizations must shift from patch-based defense toward resilient identity infrastructure and real-time threat detection.