AI Agent Security: The OWASP Agentic Top 10 Guide
Harness, Tuesday, November 3rd, 2026
Harness explains how to address the OWASP Top 10 for Agentic Applications across development, delivery pipelines and runtime.
This Harness blog post, based on a webinar by Harness senior director Renny Shen, argues that AI agent security is often treated only as a runtime problem.
Because agents can plan, select tools, use credentials, retain memory and act without human review, minor weaknesses can lead to consequential actions such as following hidden instructions or misusing tools.
The post maps the OWASP Top 10 for Agentic Applications, published in December 2025, to the software lifecycle to show which risks to address in development, which controls to enforce in pipelines and which require runtime protection.
Its takeaway is that runtime protection catches what scanning misses, while scanning reduces what runtime protection must catch.