Agentic Pentesting
Definition
Agentic pentesting is penetration testing carried out by AI agents that take actions, observe the results and choose what to do next.
What Is Agentic Pentesting?
In agentic pentesting, specialized AI agents divide the work of a pentest: one maps the attack surface, others probe specific vulnerability classes, and a coordinating layer shares what each one learns. Some vendors pair the agents with human testers who validate results.
Components of Agentic Pentesting
- Model: the reasoning engine, general-purpose or trained for offensive security
- Harness: the tools, browser, proxy and orchestration the agents use
- Memory: what the agents have learned about the target
- Validation: how findings are confirmed before they are reported
Agentic Pentesting Examples
An agent logs in as a standard user, notices an order ID in an API call, and tries other IDs. When it retrieves another customer's order, it records the request, the response and the steps, and hands the finding to validation.
Related ranking
Related terms
AI Red TeamingAI red teaming is adversarial testing of AI systems, such as models, LLM applications and agents, to find ways they can be manipulated or abused.Proven ExploitabilityProven exploitability means a vulnerability has been confirmed by actually exploiting it, typically documented with a working exploit and reproducible steps.Business Logic VulnerabilityA business logic vulnerability is a flaw in how an application is designed to work, such as broken authorization or a workflow that can be bypassed, rather than a flaw in input handling.Exploit ChainAn exploit chain combines several weaknesses, often individually low severity, to reach an impact that none of them reaches alone.