Meaningful oversight gives reviewers information they can understand and process—not a dump of every agent action and output. Pair a human-readable reasoning summary with the full action list, then build review into the workflow. section 6.7.1
Use review controls that match the risk and pace of the workflow:
- Compare each run’s goal with the expected goal stored for it. section 6.7.1
- Auto-approve low-risk actions, escalate risky ones, and block critical ones. section 6.7.1
- Require approval before a workflow deviates from an approved plan; stop safely and escalate when needed. section 6.7.1
- Use automated judges when actions arrive too quickly for real-time review, while accounting for biased evaluations. section 6.7.1
Table 6.7 Controls for ethical agentic deployments
|
Control
|
Why it matters ethically
|
|
Classify impact of actions and tools
|
Impact estimates ensure subsequent controls
(e.g. human oversight, autonomy) activate before harm is done.
|
|
Autonomy tiers
|
Ensures that autonomy is proportional to the ethical
impacts that may harm individuals.
|
|
Accountability
|
A named human or organization is accountable for harms the
agent causes[142].
|
|
Plan preview for consequential tasks
|
Provides a clear explanation of the plan of action and
allows the reviewer to make changes to the plan to prevent harms.
|
|
Human oversight
|
Provides meaningful and accessible oversight, built around
what a reviewer can actually process.
|
|
Workflow design
|
Checkpoints and constraints outside the agent’s control
keep the workflow from drifting off its goals. Graceful failure design
ensures a halted run causes no new harm.
|
|
Redress path
|
Protects people from being trapped inside an automated
process and aligns with GDPR and EU AI Act oversight requirements.
|
|
Monitoring and logging
|
Collects the evidence needed to redress harms and prevent
recurrence.
|
|
Multi-step testing
|
Provides detailed understanding of how harms may propagate
in the agentic system.
|
|
Memory auditing
|
Prevents bias from accumulating across sessions and
bending future runs off their intended goal.
|
The control table places human oversight alongside impact classification, autonomy tiers, plan previews, and workflow safeguards. section 6.7.1
