Working Model for AI Safety begins with a simple engineering premise: an AI-enabled safety system should not act merely because it can sense, classify, or calculate. It should act only when requirements, Usecases, operating envelopes, knowledge sets, and verification evidence have already defined what the system is allowed to do.

AI-enabled safety systems are often discussed as if perception alone creates capability.

That framing is incomplete.

A vehicle system does not gain authority simply because it can sense, classify, or calculate. Authority must be engineered before execution occurs.

In complex safety-critical systems, the Working Model becomes the boundary between what the system can perceive and what it is permitted to do. It defines the relationship between requirements, Usecases, operating envelopes, targeted knowledge, verification, and execution.

The purpose of Usecases is not merely to document scenarios.

Usecases are the engineering bridge between requirement intent and executable safety behavior. They define what question must be answered, under what conditions, with what knowledge set, and with what proof before the function is allowed to act.

This whitepaper opens Book III of the Applied Philosophy of Systems Engineering series by examining the Working Model as a boundary for perception, verification, and control. It explains how finite Usecases allow complex AI-enabled safety functions to remain bounded, verifiable, and executable without surrendering system authority to open-ended data or algorithmic behavior.

The central argument is practical:

Requirements define intent.
Usecases define executable scope.
The envelope defines valid conditions.
The knowledge set supports interpretation.
Verification must occur before execution.
Technology must then execute fast enough inside that bounded authority.