Graceful failure limits harm when an agent encounters a problem or a workflow is halted. The safe default is to block the workflow and escalate to a human, while ensuring a stopped run does not leave work half-processed. section 6.7.1
When an agent behaves unexpectedly, containment can preserve limited operation while restricting its ability to cause harm:
- Require more human approvals, reduce available tools, or narrow permitted actions while operators investigate. section 4.8.7
- Use workflow checkpoints and policy rules to block unsafe actions and escalate halted runs to a person. section 6.7.1
- Keep a kill switch that can halt operations, revoke credentials, and isolate the agent; support both automated triggers and manual override. section 4.8.7
A claim halted mid-process
A stopped claims workflow should not leave a claim stuck in limbo or half-processed. Block the run and escalate it for human attention. 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 controls table identifies workflow checkpoints and graceful failure as safeguards against an agent drifting from its goals. section 6.7.1


