Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

WeaknessASC-2026-0043Other weakness

No human approval for consequential actions

MediumFixed

Payments, deletions, external messages and permission changes happen on the agent's decision alone.

How to detect it

  • Tools that send, pay, delete or grant without a confirmation step
  • Approval prompts that do not show what will happen

How to resolve it

  • Require explicit approval for irreversible or external actions
  • Show the exact action and data in the approval prompt

How to investigate

  1. List consequential actions taken without approval
  2. Reverse them where possible

Sources

Open it in the Learning Center's vulnerability database