Agents are rewriting the rules of security. Here’s what engineering needs to know.

AI agents can autonomously access code, tools, and external systems, greatly expanding the software security attack surface.

New risks include prompt injection, backdoors, data exposure, and destructive actions, forcing security teams to rethink the threat model.

NIST is studying how to track and evaluate these systems, since failures could affect safety, trust, and adoption.

Organizations adopting AI agents need new security controls and monitoring to prevent small weaknesses from becoming major breaches.