What This Looks Like
A tool, connector, API, integration, or runtime call fails, but the system treats the failure like a policy decision, safety block, approval result, compliance issue, or governance outcome. A technical failure may become a refusal, escalation, risk label, or policy-style answer.
Why Users Blame AI
The user sees the final AI behavior, not the low-level tool failure. The response may look like the AI decided something was not allowed, unsafe, noncompliant, or unapproved, when the original cause was a runtime, connector, schema, permission, or availability failure.
What to Check First
- Whether a tool, API, connector, or runtime error occurred first
- Whether the failure was converted into a policy-style message
- Whether the workflow distinguishes runtime failure from governance decision
- Whether retry or fallback behavior exists for tool failure
- Whether escalation rules treat missing tool data as risk
- Whether human review sees the actual failure cause
When This Is AI-Adjacent
Use this AI-Adjacent Issue when a technical tool failure is elevated into a policy or governance decision. If the risk signal escalates beyond evidence, use the related Workbench Issue. If human and automated review disagree after the failure, use the related review-disagreement Issue.