An Alabama court threw out an appeal because the brief cited decisions that had never been handed down. An AI had written them. The court sanctioned the lawyer. Nobody proposed punishing the software, and almost nobody can say why that is right.
The machines can reason. They are already sharper than the people who built them, in ways that are easy to measure, and that was never the question. The question that decides how we live alongside these systems is whether anything can be at stake for them: whether a machine can stand under a claim it did not issue and cannot dissolve.
Drawing on cases from courtrooms, clinics, benefits offices and hiring systems, Hammat shows that our instruments measure how much something is wanted, never whether it has authority over the wanting. Training installs conformity. Answerability was never in the data, and nothing in the procedures now in use has been shown to put it there.
Two things come out of the argument that a reader can use. The first is a failure mode he calls silent obsolescence: not the machine doing what nobody intended, but the machine going on doing exactly what was intended, with undiminished competence, in a world that has quietly stopped being the one the intention was written for. The second is a test for how hard any system should be held – not by how capable it is, and not by what sector it sits in, but by whether, when it goes wrong, a claim would have anywhere to land.
This is not a warning about intelligent machines. It is a design brief for the institutions that will have to hold them, and, at the last, an argument about the kind of being that can be held.