Workbench / Issues

Conflicting Instructions From Different Authorities

Instructions from different sources, roles, policies, prompts, tools, or workflow authorities conflict without a clear rule for which one governs.

What This Looks Like

The AI receives instructions from multiple authority sources that cannot all be followed at the same time. A system rule, user request, policy, workflow instruction, tool constraint, reviewer note, or role expectation may point in a different direction, and the output does not make clear which authority governs the result.

Why It Matters

Conflicting authority creates unstable behavior. The AI may follow the wrong instruction, mix incompatible requirements, ignore a higher-priority rule, or appear inconsistent across similar cases. Users then have to determine whether the problem came from the prompt, the policy, the tool, or the ordering of authorities.

Structural Signal

Multiple instruction sources occupy the same decision space without a declared precedence or reconciliation rule. The issue is not merely that the AI disobeyed an instruction; it is that the governing authority structure is internally conflicted.

Common Triggers

When to Use This Issue

Use this Issue when the AI is governed by conflicting instructions from different authorities and the conflict affects the output, action, routing, or refusal.

When Not to Use This Issue

Do not use this Issue when a single instruction is vague, missing, or simply ignored. Do not use it when the authorities are compatible but the AI makes a mistake. This Issue applies when the instruction set itself contains an unresolved authority conflict.

Category

Rules & Policies

Primary Pattern

PAT-0100 — Authority Collision

Declared Patterns

Derived Primary Lenses

Derived Secondary Lenses

Search Intents

Ontology Metadata

Code
ISS-0035
Version
ISS-0035@0.1.0
Ontology release
0.1.0
Updated
2026-05-10T00:00:00Z

History

No public history entries recorded.

View full ontology changelog