Long before LLMs, the NHTSA Levels of Automation framework mapped how authority is shared between a human and a machine. The scale runs from the computer offering no assistance to the computer acting fully autonomously, with many intermediate steps where the human decides, the computer decides, or both collaborate under defined conditions.
The framework is useful because it does not ask how smart the machine is. It asks who decides, who can override, and what happens when the situation changes. Those are the same questions that define the enforceable and reproducible dimensions of the VERIFY rubric today.
Original paper
The PDF is embedded below. If your browser does not render it, use the download link.
The NHTSA levels of automation are not a maturity ladder to climb blindly. They are a design vocabulary. The right level for a given task depends on the cost of error, the speed of the loop, and whether a human can still intervene meaningfully. Every autonomy project should start by naming the level, not by assuming it should be as high as possible.