Berkeley Vulnerability Initiative · Agentic Vulnerability Coverage Map

A coverage map of vulnerabilities discovered by agentic systems, rebuilt daily from public CVE feeds.

An open record of the CVEs discovered by agentic systems, updated daily, built to surface research gaps, identify regressions in specific vulnerability classes, and map where agentic security work is gaining ground. Headline findings We compare the past 90 days (May 20 to Aug 18, 2026) against the same window in 2025: 332 identified agentic CVEs, 24,083 remaining 2026 filings, 10,230 CVEs in 2025. Web regressions 1 −54% fewer web findings (XSS, SQLi, CSRF), against a rising CVE total -628 fewer XSS, SQLi, and CSRF CVEs in our 2026 window, even as total CVEs grew by +14,185 (+138.7%).…

saved by

• Elisa Zhang

related reading

• FIRST Mid-Year Vulnerability Forecast Confirms Historic Surge, Projects ~66,000 CVEs in 2026first.org
• Frontier AI Cybersecurity Observatorycybergym.io
• GitHub - requie/AI-Red-Teaming-Guide: A comprehensive guide to adversarial testing and security evaluation of AI systems, helping organizations identify vulnerabilities before attackers exploit them.github.com
• Emergent Cyber Behavior: When AI Agents Become Offensive Threat Actors - Irregularirregular.com
• Akritesakrites.org
• Security incident disclosure — July 2026huggingface.co
• Introducing MAI-Cyber-1-Flash inside MDASH | Microsoft AImicrosoft.ai
• Vulnerability Research Is Cooked - Quarrelsomesockpuppet.org
• Assessing Claude Mythos Preview’s cybersecurity capabilities \ Anthropicred.anthropic.com
• Project Glasswing: Securing critical software for the AI era \ Anthropicanthropic.com
• Mythos finds a curl vulnerability | daniel.haxx.sedaniel.haxx.se
• LLM-discovered 0 days \ Anthropicred.anthropic.com