Glossary
Human in the loop
A person approves or rejects an agent's sensitive actions before they happen: the run pauses at the tool call and continues only on that person's answer.
OWASP recommends explicit approval for high-impact or irreversible actions. The answer has to come from an authenticated person: the OpenAI Agents SDK documentation notes that holding a run's id is not authorization. Keep approvals for the actions that matter, so each one gets a real look.