Enforce approval before the tool can submit: The team proposes relying on a system instruction saying, “Do not
Model instructions are not authorization boundaries; enforce approval before the agent can submit a benefit change.
The question
An EU public-benefit agent can search case records and submit benefit changes. The team proposes relying on a system instruction saying, “Do not change benefits without approval.” Which control must be enforced before a consequential change?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Train on approved-change examples.Examples may influence behavior, but they do not reliably enforce approval or permission boundaries during live cases.
- Review a sample of submitted changes monthly.Sampling may identify problems retrospectively, but it does not prevent an unapproved consequential change.
- Enforce approval before the tool can submit. ✓The submission tool should require an external approval or equivalent enforced authorization before a consequential benefit change is executed.
- Tune response randomness.Temperature affects output variability, not whether the agent has permission to submit a benefit change.
The trap
For consequential actions, require authorization at the tool or workflow boundary, not only in the prompt. How to remember it
Model instructions are not authorization boundaries; enforce approval before the agent can submit a benefit change.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Deployment and Use questions
- Minimize and filter retrieved context before generation: What additional control is most important? →
- The model may confidently propose unsupported routes: Which residual risk still most specifically requires →
- Reviewers cannot intervene effectively: Which remaining workforce-readiness risk is most decisive? →
- All 424 Understanding How to Govern AI Deployment and Use questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Deployment and Use ·
Every answer, right and wrong, comes with its own explanation.