Agentic AI – What Security Risks Comes With Giving An Agent A Tool Access & How Do You Mitigate Them?
Table Of Contents:
- Prompt Injection (Unintended Tool Calls)
- Over Privileged Tools (Excessive Scope)
- Unvalidated Tool Inputs(Injection At Tool Layer)
- Runaway Loops.
- Data Exfilteration By Tool Chaining.
- Credential/Secret Leakage.
- Tool Poisoning.
- Summary Table.
- Thumb Rule.
- Security Measures.
(1) Prompt Injection
(2) Over Previleged Tools
(3) Unvalidates Tools Inputs
(4) Runaway Loops
(5) Data Exfiltrations
(6) Credential/Secret Leakage
(7) Tool Poisoning
(8) Summary Table
(9) Thumb Rule
(10) Security Measures
