Definition
Security practice replacing periodic manual penetration tests with always-on agentic systems that continuously probe attack surfaces, chain vulnerabilities, and validate exploitable kill chains — simulating real adversary behavior at machine speed.
Key Points
- 2026-10: armadin exemplifies category — agent swarms produce decision-grade proof (reachability, blast radius) vs. stale quarterly pentest reports (2026-10-02-armadin-series-b-securityweek)
- Driver: AI lowers attacker cost to discover and chain weaknesses; defenders need offense-trained automation to keep pace (kevin-mandia quote)
- Contrast: Traditional scanners assess vulnerabilities in isolation; agentic swarms combine minor issues into validated attack paths