Mode and permission are different questions
Operating mode classifies behavior: what the system does in the work, determined by the work rather than by its label. Functional Permission states the allocation: what the performer may do without prior human intervention. The deposited definition:
The permission under which an AI performer operates at a workflow position: the data it may access, sense, or filter, and the actions it may recommend, initiate, execute, or coordinate without prior human intervention, as allocated by the organization.
Functional Permission · Appendix A, Revision 6.0.1 · DOI 10.5281/zenodo.22072938
A single AI performer may operate in different modes at different workflow positions, and a product's marketing name settles neither question. A product purchased to answer questions may operate as an Orchestrator when it is permitted to plan, browse, and invoke tools before responding.
Walk the modes
Each panel carries the deposited definition, the control implication, and an example of the mode in the work.
What the modes are not
The modes are non-hierarchical. They are not a ladder, not a maturity sequence, and not a ranking of risk. As delegated behavior changes, the control that must be designed changes with it: gating output, reaching what was filtered, moving control upstream into permissions and constraints, protecting the trajectory, or reaching the interactions among systems.
Ensemble is a deployment-level mode. It adds a level of analysis without replacing the classification of its members. Embodied AI is excluded from the taxonomy: physical interaction and machine agency is the first of the five hazards, not a sixth mode.
Mode does not set performer status. The performer threshold is functional and is stated on the Second Performer page.