What is AI agent security?
AI agent security is the practice of protecting software agents that use a language model to plan and take actions through tools, and of protecting the systems those agents can reach. It covers which agents exist, what each one can access, how each one can be manipulated, and what each one did.
Also called agentic AI security, agent security. Last reviewed .
Why it matters
A chatbot produces text for a person to read. An agent produces actions: it reads records, calls APIs, writes code and changes systems with the credentials it was given. OWASP published the OWASP Top 10 for Agentic Applications for 2026 on 9 December 2025 because the risks of acting systems differ from the risks of answering ones.
Attackers already treat installed agents as tools they can borrow. On 26 August 2025, malicious versions of several Nx packages were published to npm. According to the Nx postmortem, the post-install script searched the developer's machine for secrets and tried to use the AI command-line tools already installed there to help.
Related terms
- Agentic AI
- Agentic AI is the kind of software being secured.
- Prompt injection
- Prompt injection is the most common way an agent is manipulated.
- Excessive agency
- Excessive agency decides how much damage a manipulated agent can do.
- AI agent inventory
- An inventory comes first: an agent nobody listed cannot be secured.
How Guard0 handles it
Guard0 lists every agent across laptops, business platforms, and the code and cloud accounts your teams build in. It names a person who answers for each agent, shows what each one can reach, tests agents against the attacks described in this glossary, and records what they do while they run.
Read the guide: The Complete Guide to Agentic AI Security in 2026
Sources
- OWASP Top 10 for Agentic Applications for 2026. OWASP Gen AI Security Project.
- s1ngularity postmortem. Nx.
- AI Agent Standards Initiative. NIST.