OWASP’s agentic application risks catalog the most common security failures in tool-using, autonomous systems.
Why It Matters
Security for agents is not just about unsafe text. The risks are often architectural: tool misuse, identity abuse, memory poisoning, and cascading failures.