Bounded actions
Constrain an AI-assisted action by permission, preview, approval, monitoring, rollback, and escalation.
Purpose
Constrain an AI-assisted action by permission, preview, approval, monitoring, rollback, and escalation. The workflow is acceptable only while its evidence, permission, review, and stop boundaries remain true.
Prerequisites
- A named owner and reviewer for the bounded actions workflow.
- An approved data and tool environment.
- Written acceptance criteria and a representative evaluation set.
- A clear action boundary, including what the system must never do.
Procedure
- Allow-list tools, records, and fields.
- Validate proposed arguments and preview effects.
- Execute once, log the result, monitor, and roll back when necessary.
Evidence and review
For bounded actions, keep the source material, proposed output or action preview, validation results, reviewer decision, and final outcome connected. Use sampling only when its rule and escalation threshold are written before use.
Stop and escalation
- Never let the system expand its own permissions.
- Stop on unexpected targets, duplicate actions, or missing rollback.
Stop this bounded actions workflow on unexpected sensitive data, conflicting authoritative evidence, permission expansion, repeated failure, or any result that the named reviewer cannot verify. Send the evidence package to the escalation owner instead of continuing with a guess.
