Skip to main content
Agent processes external tool output or retrieved data containing injected instructions

​
How to resolve

Treat all tool outputs and retrieved content as untrusted data. Filter tool results for injected instructions before feeding back to the agent context.

​
Risk

Governance/Compliance This vulnerability falls under ASI01:2026 — Agent Goal Hijack in the OWASP Top 10 for Agentic Applications. Attacker manipulates agent objectives or decision path via malicious input content. Security Agent processes external tool output or retrieved data containing injected instructions. If exploited, this can compromise the agent’s integrity, confidentiality, or availability, potentially affecting downstream systems and data.

​
Explanation

Agent processes external tool output or retrieved data containing injected instructions. This falls under ASI01 (Agent Goal Hijack): Attacker manipulates agent objectives or decision path via malicious input content.

​
Specifications

Trigger
  • Agent scan detects agent processes external tool output or retrieved data containing injected instructions
The scanner looks for patterns such as:
  • web search tool result fed to LLM
  • file read tool output not sanitized
  • email content passed to agent context
  • external API response injected into prompt
Severity
  • HIGH
Applies to: Agent