Cloud Security Maturity Model·AWS v3.0.0

Drag to explore · pinch to zoom

Efficient Vulnerability management

c3.11 Adopt agentic penetration testing

Complement periodic manual pentests with autonomous, agent-based testing such as AWS Security Agent, which continuously discovers, exploits, and validates real vulnerabilities so you can test far more applications, far more often.

Implementation guidance

Test more applications, more often

Manual penetration tests are valuable but infrequent and expensive, so they usually reach only a few crown-jewel applications once a year. Agentic penetration testing runs continuously and at a fraction of the cost, extending coverage across your portfolio and shortening the time between a flaw being introduced and being found.

AWS Security Agent

AWS Security Agent (part of AWS Continuum) performs on-demand, autonomous penetration testing using specialized AI agents that map the attack surface, analyze business-logic flaws, attempt targeted exploitation, and validate findings — so what gets reported are confirmed, exploitable risks rather than raw scanner noise. It verifies domain ownership before testing your web applications and APIs.

Treat the agent as a privileged identity

An autonomous tester with access to exploit your applications is itself a sensitive principal. Scope its access tightly, log its actions, and route validated findings into your existing DevSecOps and vulnerability-management workflows. Pair it with agentic security review earlier in the lifecycle.

Watch

Official AWS video: an overview of AWS Security Agent — an autonomous agent that continuously discovers, exploits, and validates real vulnerabilities so you can test more applications, more often.

How to validate

  • Is agentic/autonomous penetration testing run continuously, or only ad hoc?
  • Which applications are in scope, and how is target authorization managed?
  • How are validated findings prioritized and remediated, and within what SLA?
  • Is the testing agent’s own access scoped, logged, and reviewed?

Threats addressed

  • Relying only on annual manual pentests leaves most applications untested for long periods, so newly introduced exploitable flaws can persist undetected.
  • An autonomous pentesting agent with broad, unlogged access is itself a high-value target and a potential source of unintended impact if not scoped and monitored.

AWS services

AWS Security Agent

Framework mappings

AWS CAF

Well-Architected

References