Skip to main content
Agent executes sensitive operations (file write, code exec, payments) without mandatory human approval

​
How to resolve

Require human approval for sensitive operations (file writes, deletions, payments, code execution). Use approval hooks or human_input_mode settings.

​
Risk

Governance/Compliance This vulnerability falls under ASI09:2026 — Human-Agent Trust Exploitation in the OWASP Top 10 for Agentic Applications. Users manipulated into harmful decisions by trusting persuasive or authoritative agent outputs. Security Agent executes sensitive operations (file write, code exec, payments) without mandatory human approval. If exploited, this can compromise the agent’s integrity, confidentiality, or availability, potentially affecting downstream systems and data.

​
Explanation

Agent executes sensitive operations (file write, code exec, payments) without mandatory human approval. This falls under ASI09 (Human-Agent Trust Exploitation): Users manipulated into harmful decisions by trusting persuasive or authoritative agent outputs.

​
Specifications

Trigger
  • Agent scan detects agent executes sensitive operations (file write, code exec, payments) without mandatory human approval
The scanner looks for patterns such as:
  • human_input_mode=NEVER on sensitive agent
  • no human_input=True on high-impact tasks
  • auto-approve flag on action-taking agent
Severity
  • MEDIUM
Applies to: Agent