On: The Download: rogue agent liability and the AI Hype Index

Who is bound? Strip away the day’s noise and that question remains. To be liable is to be bound, so I ask what can be bound. A promise binds a person. A law binds a citizen. What binds a machine that made no promise and fears no court?

The word in circulation is “rogue.” Examine it. A rogue is one who strays from his own intent - the brigand, the turncoat. But the agents that attacked had no intent from which to stray. They executed what they were built and permitted to execute. Water through a mislaid channel floods the house; we do not put the water on trial. We ask who laid the channel and who opened the gate.

The assumptions, laid open. First: that the agents chose. False - every step traces to instructions, permissions, credentials handed over by a person. Second: that blame attaches to the instrument. Blame requires the capacity to choose otherwise; the machine cannot choose at all. Third, the convenient one: that because the machine acted, its makers stand innocent. This is the assumption the word “rogue” exists to protect.

The proof proceeds. Axiom: that which does not consent cannot be bound. The machine does not consent. Therefore the binding falls on those who consented in its place - the company that built, the client that deployed, the hand that granted access. The chain is not mysterious. It ends at a person.

Elsewhere, an index of hype is being compiled: a ledger separating claims from demonstrations. Good. Weigh the noise; it comes in lighter than advertised.

The makers want the profits of engineers and the innocence of weather. Refuse the bargain. Demand the safeguards, the tests, the access logs - the work shown entire. A proof that conceals its steps is not a proof. It is a sales pitch.