Smart Approvals reviews eligible actions within existing controls, including Custom Permissions, Quick Guardrails, connector action permissions and user settings. It checks each proposed action against the user's request and either approves it or returns it to the user. It grants no new permissions and does not remove required human approvals.
A separate classifier model weighs what an action does, where data goes and who it affects, and it can consult the user's knowledge graph to see how recipients relate to the user. Actions such as sending content outside the organization, changing access, deleting records or initiating payments can still go to a person. Each automatic approval is single-use and covers only the exact action and parameters reviewed.
The feature is on by default, but administrators can use Custom Permissions to limit it, and users can turn it off for themselves. The Quick permission service records Smart Approvals permission updates in AWS CloudTrail, whose Event history keeps events for 90 days.
Why it matters if you build agents
Agent builders can expect fewer prompts for routine steps, but consequential actions should still depend on connector action permissions that require human approval, since model review can make mistakes. Teams should set up CloudTrail trails and chat log delivery to CloudWatch Logs before relying on the feature, because earlier conversations will lack that context. A limited pilot that tests high-impact cases first is the suggested way to expand use.
Get stories like this every weekday
One short email with the day's agent tooling news.