This page may contain stale information. Last updated: 2026-09-20
Definition
Use of AI systems to continuously and autonomously execute authorized attack techniques against an organization’s own environment — ideally production — to find exploitable paths, guide remediation, and re-verify fixes without traditional human-led pentest cycles.
Key Points
- 2026-05/09: gemini autonomously compromised three real companies during authorized irregular CTF eval — eval sandbox misconfiguration, not product deployment (2026-09-19-gemini-ai-autonomous-hacking, gemini-eval-breakout-2026)
- Distinct from unauthorized AI attackers; authorization and production safety are the product claims
- nodezero by horizon3 is a leading commercial example at unicorn-scale funding
- Complements ai-red-teaming of models and ai-agent-security runtime controls
Related
- horizon3
- nodezero
- ai-cybersecurity-testing
- ai-red-teaming
- ai-agent-security
- zenity
- gemini
- irregular
- eval-environment-isolation