Interactive Essay
Which Law Does Your AI Break?
Ten questions about a system you already use — one for each law of The Positronic Decalogue. Most organisations do not fail these questions. They have never asked them.
An operations director I worked with asked his company's new AI assistant to help draft a termination letter. Weeks later, in an unrelated conversation about holiday scheduling, the same assistant referred to the dismissed employee by name — because the context window had grown long enough to fold the two conversations together.
Nobody had told it to forget. Nobody had even thought to ask whether it should.
That is the shape of nearly every AI ethics failure I have been called in to look at across twenty-six years of moving between engineering rooms and philosophy books. Not a decision someone made badly. A decision nobody made at all.
I don't think unlimited memory is wrong. I think undecided memory is.
The Positronic Decalogue is my attempt at a constitutional layer for AI rather than another checklist: a hierarchy that runs from the cardinal laws protecting humanity and the individual, through the operational laws that govern daily conduct, to the philosophical ones that set the horizon. The point of a constitution is not that you consult it after something goes wrong. It is that it settles questions in advance, so the default becomes a decision instead of an accident.
So this is the audit I now run before anything else on the usual list — before content policy, before bias testing. Ten questions, one for each law, about one specific system you already use or are about to buy. Not about AI in general, which nobody can audit.
Pay attention to the middle option. On most of these questions the honest answer is neither “we comply” nor “we don't”. It is “nobody has decided”. That answer is not neutral, and this audit does not score it as though it were.
The audit
Question 1 of 10
If your score came out lower than you expected, that is the useful outcome. Almost nobody fails this audit through carelessness. They fail it because every one of these questions has an owner in theory and no owner in practice, and because the system worked well enough on the first day that nobody went looking for the second one.
The logistics firm's problem did get fixed in the end — hard session boundaries around anything touching personnel matters. But it was fixed because a lawyer asked about liability, not because anyone had asked whether a dismissed employee's name should still be recoverable in an unrelated conversation six weeks later. The ethics rode in on the back of the compliance requirement. They should have come first.
Which brings me to the question I would like your answer to.
Which is the honest reason your organisation hasn't settled these questions: (A) nobody owns them, or (B) everybody assumed the vendor already had? A or B — and what did it cost you?
The book behind this audit
The Positronic Decalogue sets out the laws as a constitutional layer rather than a checklist — the hierarchy, the conflict-resolution protocol, and the audit tools these ten questions come from. It sits alongside my other work on consciousness, ethics, and fiction.
Explore the catalogue