Workbench / AI-Adjacent Issues

Guardrail Applies to One Model Path Only

A guardrail, policy, validation rule, safety check, or workflow constraint applies to one model path, mode, runtime, or route but not another.

What This Looks Like

A guardrail, validation rule, safety check, policy constraint, approval gate, or workflow rule applies in one path but not another. The same task may be checked in one model, mode, route, runtime, agent path, or product surface, but bypassed or applied differently elsewhere.

Why Users Blame AI

The behavior appears inconsistent across AI paths. The user may see one route block, validate, warn, or constrain the task while another route allows it or handles it differently. The underlying cause may be routing, model selection, product mode, runtime configuration, or guardrails wired to only part of the system.

What to Check First

When This Is AI-Adjacent

Use this AI-Adjacent Issue when a guardrail applies to one model path or runtime route but not another. If the result is unexplained routing or inconsistent outcomes between similar cases, use the related Workbench Issues for routing divergence or excessive variance.

Common Ways People Describe This

Metadata

Code
ADJ-0016
Layer
ai_adj_issue
Version
0.1.0
Updated
2026-05-10T00:00:00Z