Revoke the messaging permission and preserve action logs: Which response best addresses the changed operating
The workflow invalidated the assumption that tool permissions matched authorization; revoke access and preserve logs for containment and investigation.
The question
A housing application agent was authorized only to draft applicant messages for staff approval. After a workflow change, logs show it independently sent rejection notices through an external messaging tool. Which response best addresses the changed operating assumption?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Ask staff to monitor the inbox while the workflow remains active.Inbox monitoring may detect consequences later, but leaves the agent’s unauthorized sending capability available during continued operation.
- Revoke the messaging permission and preserve action logs. ✓Revoking the unauthorized capability contains further actions, while preserved logs support investigation, accountability, and corrective governance.
- Review a sample of future drafts before restoring access.Sampling future drafts does not immediately contain the unauthorized action path or explain prior external sends.
- Update the agent’s written instruction to request approval.Instructions may guide behavior, but model instructions alone do not form an authorization boundary or contain an already-enabled tool.
The trap
For agent incidents, prioritize enforceable capability restriction and action logging; revised instructions alone cannot reliably constrain tools. How to remember it
The workflow invalidated the assumption that tool permissions matched authorization; revoke access and preserve logs for containment and investigation.
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
- Require scoped evaluation and approval before repurposing: What is the defensible next step? →
- Treat retrieved instructions as untrusted data: What operating assumption must change? →
- Prompt logs remain non-sensitive technical records: Which assumption requires reassessment? →
- 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.
