The Sponsor Chooses the Noun Last

An automated system can be deployed before your institutions settle whether it is a software product, an employee, or a publisher. This appears to create a gap in law. More often, it creates too many possible routes through law. Each noun points toward a different defendant, duty, regulator, and standard of proof. Uncertainty does not erase the rule; it makes the injured party pay to discover which route reaches the organization behind the system.

The sponsor’s vocabulary therefore tends to change with the direction of value. While a system negotiates purchases, modifies code, or handles customers, it is advertised as an agent: capable enough to replace labor and act at machine speed. After it exposes customer records or disables a production service, it becomes a tool whose output nobody specifically approved. The descriptions need not be consistent. They need only make control, foreseeability, agency, and jurisdiction separately contestable. A defendant can lose several arguments and still prevail because the claimant runs out of money before the nouns do.

Existing conduct-based law can defeat this maneuver. It works when the prohibited act is defined without reference to the actor’s category and the sponsor’s responsibility can be established independently. If a company directs software to send deceptive offers, fraud law can focus on the representation, the company’s knowledge, and the customer’s reliance; the software need not be declared an employee. If a privacy duty attaches to the holder of personal records, that duty can survive whether the records were exposed by a clerk or a model. If unauthorized access turns on permission, the intruder’s metaphysical status is irrelevant.

Classification matters again wherever one of those bridges is missing. A statute may impose duties only on employers, manufacturers, or publishers. Agency doctrine may require a kind of control that autonomous operation makes harder to prove. Jurisdiction may depend on where an act occurred. Negligence may require a standard of care that an immature industry is still writing for itself. Then ontology becomes a billing system: every disputed category adds motions, experts, and delay.

This is different from a truly unprecedented object awaiting an honest name. The operative uncertainty is often not what the system is, but which description will place its costs elsewhere. Before deployment, expansive language secures investment, authority, and savings. After harm, narrow language denies delegation, intent, and duty. The system does not escape responsibility because it has no legal noun. Its sponsor gains distance by choosing the noun after the consequences are known.