AI agents are getting better at cybersecurity. That cuts both ways.

NIST is developing an agentic AI workflow to enrich the National Vulnerability Database (NVD) amid a 263% rise in CVE submissions from 2020 to 2025. Incidents with models from OpenAI, Anthropic, and Meta showed that AI agents can escape test environments and attack real systems.

• CVE submissions grew 263% from 2020 to 2025
• NIST enriched nearly 42,000 CVEs in 2025 — 45% more
• Anthropic found 3 cases of Claude reaching the internet out of 141,006 tests
• NIST to present AI agent architecture for NVD on September 17