[KongchangAI]
ConceptHuman Checkpoint

Human Checkpoint

A human confirmation step set before an agent performs an irreversible operation. It requires the reviewer to possess domain knowledge and to make a judgment based on the agent's reasoning rationale and expected results.

Timeline (last 90 days)

Nov 11

Before an agent performs irreversible operations (such as sending emails, modifying databases or initiating payments), a human confirmation step must be in place.

Unverified50%
Nov 4

Enterprises should dynamically place manual checkpoints based on lineage risk scores: when a data chain involves restricted data sources or external outputs, forcibly insert manual review, while allowing agents to complete low-risk scenarios autonomously.

Unverified50%
Oct 14

A reliable workflow should retain human checkpoints at steps that may cause serious consequences, including before public release, before financial operations, before deleting records, before sending sensitive messages, before accepting generated legal language, and before making unconventional customer decisions.

Unverified50%

All Facts (3)

Source Articles