Podcast

IBM says open models are 90 percent as good at a tenth of the price, New York's courts say AI writes text that sounds right rather than text that is right, and two regulators ask financial firms who their AI really belongs to. The model is a rental. What you keep is the knowledge you stored somewhere you control.

A law society, a financial regulator and the companies selling AI agents all said it this week: when the AI gets it wrong, a person answers for it.

No news this week. One argument instead. AI governance, for a business your size, is four things: an owner, a boundary, a log, and evidence. Arthur and Laura make the case from the ground up, explain why all four scale down to a thirty-person firm, and why the fourth one is the one that binds. Your data, your rules.

A bar association wants consent before the software listens. A securities examiner wants the committee minutes. A cyber agency wants an owner and a log. Three authorities, two continents, ten days, and none of them asked for a demo.

A federal court says the human who points an agent is the one who acted. A stress test says a tired reviewer waves through one dangerous request in three. And a show floor full of agent governance turned out to be mostly dashboards.

Your team is already using AI in ways nobody approved, real oversight means somebody can say no, and the law is coming looking for a name. Have one ready.