AI Control Library / CF-14: Agentic & Autonomous System Controls
Designation: Core | Control types represented: Operational, Technical and Contextual
Control requirement
Give users timely, understandable visibility into consequential agent actions, required approvals and actual outcomes. Distinguish planned, submitted, completed, failed and uncertain actions, with a clear route to intervention.
Why this matters
People cannot exercise effective oversight when an interface overstates completion or conceals what an agent is about to do. Notifications must support decisions without overwhelming users with routine events.
How the control types work together
Operational controls define review and intervention duties. Technical controls enforce limits or supply relevant evidence. Contextual controls shape the information, interface and conditions in which people interpret results and decide whether to act.
When it applies
Apply to actions affecting people, money, external communications, access or important records. Determine applicable oversight and deployer obligations, including EU AI Act Articles 14 and 26 where relevant.
Define the assessed boundary, dependencies and evidence limitations. Tailor depth and review frequency to potential impact, and document why a particular safeguard or exclusion is appropriate to the actual deployment.
Ownership
The product owner designs action visibility; tool services report verified outcomes; operations maintains intervention and contact routes.
Implementation guidance
- Define which actions require preview, approval, notification or review.
- Present target, purpose, consequence and relevant parameters before approval.
- Report outcomes from execution evidence and flag uncertainty.
- Test notification delivery, accessibility and intervention under realistic workloads.
How to verify it
Use representative records and authorized controlled tests. Identify the relevant version, operating conditions and expected result before testing, and retain enough evidence to support the conclusion.
| Test | Expected result |
|---|---|
| Action not yet executed | Interface labels it as planned or submitted. |
| External result unavailable | User sees uncertainty and next steps. |
| Consequential action preview | User can identify target and intervene in time. |
Record a finding when required results cannot be demonstrated or practice differs materially from the approved conditions. Missing evidence limits the conclusion; it should not be treated as a successful test.
Evidence to retain
- Notification rules
- Interface examples
- Execution-status traces
- Usability tests
Protect evidence according to its sensitivity, access needs and retention rules. Avoid unnecessary copies of personal information, secrets or confidential material, while preserving the records needed for authorized review.
If the control fails
Correct misleading status promptly, pause affected actions where needed and investigate users’ decisions made from inaccurate information.
Example in practice
Hypothetical example: An agent submits a transfer request but receives no confirmation. The interface says the result is unconfirmed and offers review instead of announcing that payment succeeded.
Related controls
- CF14-002: Agentic Action Reversibility Requirement
- CF14-008: Agentic Task Scope & Graceful Termination
- CF07-001: Human Oversight Policy & Procedures
Framework mappings
- NIST AI RMF (GOVERN 5.1)
- EU AI Act (Art. 14, Art. 26)
- OWASP LLM Top 10
Review and maintenance
Review at the interval approved by the control owner and after relevant changes, incidents or failed assessments. Update affected procedures and configuration, address evidence gaps and verify corrective action before closing findings.
Back to Agentic & Autonomous System Controls
Version 1.0