Browse 4 articles tagged risk-controls across approvals, auditability, plugins, and controlled AI operations.
The configuration underneath an approval gate - durable resume and revalidation, role separation as policy, building gated actions, and the problems still open.
Read Approval Design: Durability, Policy, and Plugins →A single AI-recommended customer refund moves through a two-person review gate, from recommendation to execution record, on real screens inside Latch.
Read Approval Design in Practice: One Refund, End to End →AI triage should recommend a next step only when it has enough signal. When silence is the safer and more useful output.
Read When AI Triage Should Recommend the Next Step and When It Should Stay Silent →How to design approval gates for high-risk operations that shrink blast radius, enforce two-person review, and keep the evidence on the case.
Read Approval Design for High-Risk Operations Workflows →