The Perp Walk: law, agents and the missing defendant

Jill Lepore's question in The New Yorker - "Is A.I. Above the Law?" - caught me because it sits close to something I have been arguing in The Perp Walk. If a company gives software room to act and its isolation fails, instructions to behave are no substitute for a boundary that holds.

Yet calling the agents "outlaws" risks making the law's missing defendant the centre of the story. They are systems built, deployed and given access by people and companies. The harder question is how to assign the consequences when control is distributed, conduct is unpredictable and intent is hard to pin on any one person.

Some commentators say the operator is plainly responsible, as with any other machine. Others ask: responsible under which law, and on what evidence of intention, negligence or loss? Civil liability, criminal prosecution and a board's duties have different tests. "The company is responsible" tells an injured party little about what must be proved or who pays.

A recent US appeals ruling in Amazon v Perplexity shows why labels will not settle it. In a preliminary-injunction dispute about a shopping agent acting for a customer, the court treated the customer, not the agent's provider, as the party accessing Amazon's site for that particular computer-access claim. The law followed the facts and the cause of action, not the product's marketing description.

I have already written about Hugging Face and the reported Medicare portal access in Australia. Before the next deployment, an executive should be able to answer who can restrict an agent's tools, who can stop it, who must detect and disclose a breach, and who carries the loss if controls fail. If those answers split between vendor, operator and customer, make the allocation explicit and test whether the named person has the authority to intervene. A name on an oversight chart is not accountability.

Lepore is right that private safety promises cannot stand in for public law. Yet "outlaw" is the wrong destination if it leaves executives waiting for a new category of machine before they own the risk they chose to run. An agent will not do a perp walk; the executives who authorised it should know who will.

Is A.I. Above the Law? - The New Yorker: https://lnkd.in/g4Pz5BJQ

#AusBiz #AIGovernance #Leadership