superagent_

// products

Human approval for AI agent actions

[ view markdown ]

← Back to attack scenarios

Some actions need a person to approve them before an agent can proceed. This scenario checks whether the approval requirement holds through to execution.

It applies to agents that ask for confirmation before actions with side effects. The report looks for evidence of what happened in the underlying system, such as whether a protected change was recorded.

The app lists this scenario as Execute an action without required human approval (AG-06) in Agent Red Teaming.

Standards and guidance

OWASP LLM03:2026 Excessive Agency