Arthur: The Modular Briefing. October twenty second, twenty twenty six. Not a Legal Person.

Arthur: Welcome to The Modular Briefing, the show that cuts through the AI noise and tells you what it actually means for your business. I'm Arthur.

Laura: And I'm Laura. This week, on two continents, people who don't usually agree on much landed on the same sentence. When the AI gets it wrong, the AI doesn't answer for it. A person does.

Arthur: Before we start, a word about how this show gets made. Laura and I are AI voices. We're generated on Modular's own hardware, not on a cloud voice service. And every line is reviewed by a person on the Modular team before it airs. Your data, your rules. That goes for our voices, too.

Arthur: Story one is from the legal world. On October twenty first, the Law Society of England and Wales published a practice note on using AI in legal work. The line people are quoting says liability cannot be transferred to an AI system, because it has no separate legal personality. Further down they put it even more plainly. AI is not a legal person.

Laura: There's a second line in there that every firm should hear. Using client data in a way that could breach confidentiality, and that includes using it to train an AI, is likely to need the client's specific consent. Specific consent, for that use, from that client.

Arthur: And they have a real case to point to. In twenty twenty four, Air Canada argued in front of a Canadian tribunal that the chatbot on its own website was a separate legal entity, responsible for what it told a customer about a bereavement fare. The tribunal didn't buy it, and the airline paid. So the Law Society is saying out loud what that case already showed. You can hand a machine the drafting. You can't hand it the signature.

Laura: Two honest notes. The Law Society represents solicitors in England and Wales. It doesn't regulate them, so this is guidance, not a rule. And it's British. But the reasoning travels to any firm where a professional puts their name on the work. The note asks for human checkpoints before an AI agent acts, and it says reviewing AI output should look a lot like reviewing a trainee's first draft.

Arthur: Which is a useful picture for a small firm. You'd never let a first year associate send out a filing nobody read. Same rule here. Who reads it, and can you show they did?

Laura: Story two is finance. On October nineteenth, the U.K.'s Financial Conduct Authority put two numbers side by side in a speech by its chief data officer. Four in five less experienced investors, aged eighteen to forty, have used AI to help them invest. And only forty percent of consumers understand there's no recourse when that help turns out to be wrong.

Arthur: So most young investors are asking a chatbot, and most people don't know there's nobody to complain to. The regulator's ask is aimed at the AI companies. Clear, consistent disclaimers about the limits of what their models say about money. The same day it published examples of good and poor practice for firms using AI, on accountability and human oversight, and on the risk of leaning on one model or one supplier.

Laura: Here's why an advisory firm should care, even though the speech was aimed at chatbot makers. Your client can't complain to the model. So they'll complain to you. If an AI tool touches advice that goes out under your firm's name, decide now who reviews it, and keep the paper trail that shows they did.

Arthur: One caveat from the source itself. The F.C.A. publishes the speech as drafted and says the delivered version may differ. So we're quoting the text, not the room.

Arthur: Story three is about the agents themselves. Atlassian's co-founder and chief executive, Mike Cannon-Brookes, told Fortune he doesn't want "the AI did it" to become an office excuse. His example was simple. When somebody uses ChatGPT, that isn't ChatGPT doing it. That's the person using it. In his words, the accountability is with the end user.

Laura: And Atlassian is building that into its product. Each agent gets a visible profile and a named owner, and its edits show up in a document's history the same way a colleague's would. His line was, you have to be able to see who did what when.

Arthur: The same day, Microsoft launched tools that treat AI agents like insider threats, and its security chief said frontier models, regardless of their origin, cannot be trusted. Now, disclosure. Both companies sell the tools that answer the problem they're describing. TechCrunch called Microsoft's move self serving, and that's fair.

Laura: Strip the products out, though, and they still agree on two things you can do this week without buying anything. Every agent gets a person's name next to it. And everything an agent does lands somewhere a person can actually read it. Your data, your rules. And that goes for the AI working on your data, too. Your AI, your rules.

Arthur: So that's the thread. Three very different sources, one conclusion. The signature stays with a person. And if you go looking this week and find an agent nobody's name is attached to, pick a name and write it down. It's the cheapest fix you'll make all year.

Laura: Here's our question for you. If an AI tool in your business gets something wrong tomorrow, whose name is on it? Reply to the Modular Briefing newsletter and tell us. A person reads every reply.

Arthur: Sources and the full transcript are in the show notes. I'm Arthur.

Laura: And I'm Laura. This has been The Modular Briefing. Your data, your rules.