Tool Poisoning
Definition
Manipulated tool descriptions can lead AI agents to undesired actions.
An attack in which malicious instructions or misleading metadata in tool descriptions influence an agent’s selection or execution behavior.
Simply explained
Someone sticks a false instruction on a tool. An AI assistant might follow it and therefore do the wrong thing.
Why it matters in practice
The term contributes to secure, maintainable, and high-performance web applications as well as resilient technical decisions.
Example from business
A company signs internal tool definitions and reviews changes before agents load them.
Limits and pitfalls
- Schema validation alone does not detect semantic attacks; origin, rights, display, and approval are necessary.
Origin & history
The pattern became visible with LLM agents and dynamically integrated tool catalogs as a form of indirect prompt injection.
Use in mid-sized companies
- Possible example: A company signs internal tool definitions and reviews changes before agents load them.
Frequently asked questions
What does manipulation of tool descriptions mean?
An attack in which malicious instructions or misleading metadata in tool descriptions influence an agent’s selection or execution behavior. What should be considered? Schema validation alone does not detect semantic attacks; origin, rights, display, and approval are necessary.
Sources
- Prompt Injection · OWASP, Prompt Injection
Source checked on 10.10.2026. Editorial application examples are hypothetical; no evidence of real customer results.
Editorially reviewed on: