human approval
Pausing an agent before a risky action so a person can approve, edit or reject it, then resuming from saved state.
In more detail
Show the reviewer exactly what will happen, with real arguments, not the model's summary. Rate tools by risk so approvals stay rare enough to be read, record who approved what, and treat the saved state as untrusted when it comes back.
Related terms
- agent checkpoint — The saved state of a paused agent run, from which it can resume later, for example after a person approves a tool call.
- excessive agency — Giving a model-driven system more tools, permissions or autonomy than its task needs, so a fooled or mistaken model can do real damage.
- agent policy engine — Code outside the model that decides, for every proposed tool call, whether to allow it, deny it or ask a person, based on rules written as data.
Taught in
- Anatomy Of A Prompt — AI Engineering: LLM Apps, RAG & Agents
- Planning, Memory And Many Agents — AI Engineering: LLM Apps, RAG & Agents
- Agent SDKs And Human Approval — AI Engineering: LLM Apps, RAG & Agents
- Excessive Agency — AI Engineering: LLM Apps, RAG & Agents
- Working With AI Coding Assistants — AI Engineering: LLM Apps, RAG & Agents
- Capstone: The Tasks Assistant — AI Engineering: LLM Apps, RAG & Agents
References
- OpenAI Agents SDK: Human-in-the-loop
- Claude Agent SDK: Handle approvals and user input (canUseTool)
- OWASP GenAI Security Project: LLM06:2025 Excessive Agency