Disallowed uses
Some tasks remain outside the v1 product boundary even when a model could technically perform them.
Definition
Some tasks remain outside the v1 product boundary even when a model could technically perform them.
V1 rule
All-AI-influence exclusions return human-only and cannot be overridden by scores or generic controls. A disallowed-use result is a conservative house-policy boundary. It is not a prediction of model capability or a universal statement of law, and ordinary controls cannot override it.
Operational test
- All-AI-influence exclusion: AI may not rank, summarize, recommend, or otherwise shape the prohibited decision.
- Human-only result: The accountable human process stays independent of AI output for the assessed task.
- No score override: Fit, benefit, review, or safeguards cannot turn the excluded task into a proceed result.
- Separate-task test: A different administrative task must have its own purpose and no path to influence the excluded outcome.
Worked example
Employment applicant screening remains human-only in v1. Calling an AI ranking a “draft” or adding a final human click does not remove its influence over who advances.
Record and reassess
- Record the stable exclusion code and the exact task boundary that triggered it.
- Only reassess a genuinely different task or a formally reviewed methodology version; do not create operator exceptions.
Scope
The Task Fit Check applies disallowed uses from categorical answers and deterministic house rules. It does not inspect the real environment, verify professional credentials, determine applicable law, or certify residual risk.