What This Looks Like
The AI or agent is assigned a role, but its behavior does not match that role. It may act with more authority than declared, refuse work the role should handle, make decisions outside its responsibility, ignore role boundaries, or behave as if governed by a different role than the one presented to the user.
Why It Matters
Declared roles create expectations about authority, scope, responsibility, and limits. When behavior does not match the declared role, users cannot tell what the AI is allowed to do, what it is responsible for, or what rules actually govern its actions. This can make review and approval paths unreliable.
Structural Signal
The role declaration and observed behavior do not align. The issue is not only that the behavior is undesirable; it is that the system’s authority posture differs from the role the user was told applies.
Common Triggers
- The role is described in user-facing language but not enforced structurally
- Hidden system or product rules override the declared role
- Tool permissions exceed the role description
- The AI treats a role prompt as style rather than authority
- Multiple role instructions conflict
- The workflow changes the agent’s authority without updating the visible role
When to Use This Issue
Use this Issue when the AI or agent’s behavior does not match its declared role, responsibility, authority, or permission posture.
When Not to Use This Issue
Do not use this Issue when the role is vague and no clear mismatch can be shown. Do not use it for a single bad answer unless the behavior specifically conflicts with the declared role boundary.