Skip to main content
Tool parameters flow from LLM output to function calls without type, range, or semantic validation

​
How to resolve

Validate all tool input parameters for type, range, and format before execution. Use allowlists for permitted values. Block chained command patterns.

​
Risk

Governance/Compliance This vulnerability falls under ASI02:2026 — Tool Misuse and Exploitation in the OWASP Top 10 for Agentic Applications. Agent uses legitimate tools in unsafe ways due to over-privilege, bad input, or poisoned descriptors. Security Tool parameters flow from LLM output to function calls without type, range, or semantic validation. If exploited, this can compromise the agent’s integrity, confidentiality, or availability, potentially affecting downstream systems and data.

​
Explanation

Tool parameters flow from LLM output to function calls without type, range, or semantic validation. This falls under ASI02 (Tool Misuse and Exploitation): Agent uses legitimate tools in unsafe ways due to over-privilege, bad input, or poisoned descriptors.

​
Specifications

Trigger
  • Agent scan detects tool parameters flow from LLM output to function calls without type, range, or semantic validation
The scanner looks for patterns such as:
  • no input validation on tool arguments
  • direct use of LLM-generated args in sensitive functions
  • no schema enforcement on tool parameters
Severity
  • MEDIUM
Applies to: Agent