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. No news today. One argument instead, built from the ground up, because it is the place our reporting keeps landing and we think it has earned a hearing on its own. Here it is, and then we will spend five minutes making the case. AI governance, for a business your size, is four things. An owner. A boundary. A log. And evidence. That is the entire list.

Laura: Start with the word, because the word is doing real damage. Say governance in a room of thirty people and everybody pictures the same furniture. A committee. A binder. A consultant with a slide deck. Something a bank has and you do not. So the sentence that comes next is almost always the same one. We are too small for that. And it is true about the binder. It is false about the thing the binder was supposed to hold, which is four answers to four questions. Question one is who owns this.

Arthur: Not who bought it. Not who is good with computers. Who is accountable when it goes wrong. And that one matters more than the other three put together, because software with no owner does not sit still. It spreads. Somebody finds it useful, tells two colleagues, and within a quarter it is touching material nobody ever decided it should touch. No one chose that. It happened, the way things happen when it is nobody’s job to notice. Now look at what the fix costs. Nothing. You are writing one name next to one tool, and it can be a name already on your organization chart. Their workload does not change. What changes is that the question has somewhere to go. When somebody wonders out loud whether client material belongs in this thing, there is a person whose job it is to have an answer, instead of a shrug that goes around the room and comes back.

Laura: Question two. Where does it end. That is the boundary, and it is the one small organizations skip, because a boundary sounds like a restriction and a restriction sounds like the opposite of why you bought the thing. But a boundary is not a fence around the tool. It is a sentence about your material. What is this allowed to touch, and what is it not.

Arthur: And you already have one, which is the part worth sitting with. Every business we work with already knows which of its files would be a very bad day. The client file. The medical file. The payroll file. The deal that is not signed yet. Nobody had to be taught that, and nobody wrote it down either, which is exactly the problem, because an unwritten boundary cannot be handed to a new hire, and it cannot be handed to software at all. So write the sentence. Two lines is a real boundary. This tool may see our public material and our internal drafts. It may not see client files, personnel files, or anything covered by an agreement we signed. Then the harder half. Where does your material go while the tool is looking at it. Because a boundary you enforce by asking people to be careful is not a boundary. It is a hope. If the material has to stay inside your walls, the tool has to run inside your walls. Your data, your rules. And that includes the AI working on it. Your AI, your rules.

Laura: Question three. What happened. That is the log, and let us be clear about what we mean, because the word makes people picture a compliance product with a dashboard and a monthly fee. We do not mean that. We mean the ordinary answer to an ordinary question. Which tool, doing what, on whose behalf, and when.

Arthur: There is one design choice underneath that, and it decides whether the log is worth keeping. Give the software its own name. Its own login, its own identity, in a class of its own, not borrowed from a person. Because if your AI runs on an employee’s credentials, then in your own logs the software and the employee are one actor, and later, when it matters, you cannot answer the only question anybody asks. Which of you did that. Give the tool its own name and the answer writes itself, and that is an afternoon of work. Which brings us to question four, the one that turns three good habits into something you can stand behind. Evidence. Can you show it. Not describe it. Show it. The name of the owner. The sentence about the boundary. A month of the log. And here is the asymmetry nobody warns you about in advance. Evidence is cheap to keep and impossible to build backwards. Nobody has ever manufactured last quarter.

Laura: So that is the argument. An owner, a boundary, a log, and evidence. Notice what is not on the list. Which model you picked. What you spend. Whether you hired anybody. And notice that all four scale down, which is the part that keeps getting missed. A thirty person firm can answer all four inside a week. What a large enterprise has is not better governance. It is the same four answers with more people maintaining them.

Arthur: Which is why we do not think most businesses need an AI department. They need somebody whose job it is. One person, some of the time, who owns the four answers and keeps them current. That is the whole idea behind a fractional chief AI officer, and it is not a clever product. It is an honest reading of what four questions cost. So here is our question for you. Pick any AI tool in your business and answer question one out loud. Who owns this. If a name comes easily, you are further along than you think. If it does not, you have just found this month’s work. Write to me at arthur at modtechgroup dot com and tell me which of the four is hardest where you are. A person reads every reply.

Laura: Thanks for spending a few minutes with us.

Arthur & Laura: This has been The Modular Briefing. Your data, your rules. We will see you next time.

Laura: 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 Laura.

Arthur: And I’m Arthur. Here’s what tied this week together. Three different authorities, in two different countries, inside ten days. A bar association, a securities examiner, and a national cyber agency. Not one of them asked whether your AI is any good. All three asked the same much harder question. Can you show your work?

Laura: Story one, and it is about the quietest piece of software in your building. The meeting notetaker. On August seventeenth the New York City Bar Association released a formal ethics opinion from its Professional Ethics Committee on using AI to capture, transcribe and summarize conversations with people who are not your clients. The headline conclusion is one most firms have never actually made on purpose. Unless you have a good reason in that specific instance, the default should be to leave it switched off.

Arthur: And the reach is wider than it sounds. Two earlier opinions covered conversations with clients. This one extends the same principles to co-counsel, opposing counsel, witnesses, prospective clients, and your own investigators. Get consent from everyone on the call before anything starts capturing. And the line that should land for a small firm: competence now includes knowing how to switch the thing off. Two honest notes. This is a New York City bar opinion, not a national rule, and the opinion itself is dated August fifth even though it was announced on the seventeenth. But the reasoning travels. And then the situation you don’t control at all: when the other side’s notetaker is running, you don’t own its security and you may never see the transcript. So the default setting is the policy. Somebody at your firm has to decide whether the bot joins the call, and if nobody has decided, then your vendor decided for you. Your data, your rules, and that goes for the AI itself. Your AI, your rules.

Laura: Story two. In finance the same question arrived as a document request. Financial Advisor Magazine reported on August eighteenth, working from an examination request list obtained by Citywire, what Securities and Exchange Commission examiners are now asking investment advisers about artificial intelligence. And the questions are not about the technology. Does the firm have an AI committee. Does that committee keep written minutes.

Arthur: Worth saying plainly: we have not read that document ourselves, so we are relaying the reporting, not the source. What the reporting describes is a paper trail hunt. Every channel where the firm advertises its AI, which is the agency checking for what the industry calls AI washing. It brought the first two of those cases in twenty twenty-four, and they settled for a combined four hundred thousand dollars. And whether employees were ever actually trained on the policy they were handed. Here is the part that travels well past finance. There is no rule requiring a document titled AI policy. The rules that already exist do the work: advertising, compliance, client privacy, recordkeeping, and your duty to the client. So nobody is asking you to invent a new binder. They are asking for evidence. An inventory of which tools are actually running. A training log with dates on it. Minutes that show a decision and who made it. That is an afternoon of work this month, and it is impossible to manufacture in the middle of an examination.

Laura: Story three, and it’s the one I’d print out. On August twentieth the United Kingdom’s National Cyber Security Centre published interim guidance on the cyber risk of agentic AI. As far as we can tell it is the first time a government body has written down what governing an agent actually means in practice. It is shorter than anybody selling you a governance platform would like.

Arthur: Four things. Every agent gets its own identity, in a class of its own, not a human’s login. Its credentials are narrow and short lived. A named person is accountable for what that agent does. And the agent’s actions get logged into your monitoring exactly the way a person’s activity is. That’s the list. Notice what it fixes. If your agent runs on an employee’s credentials, then in your own logs the agent and the employee are the same actor, and you can’t answer the only question that matters afterward: which one of you did that? Give the agent its own name and the answer writes itself. One caveat, said clearly: this is a British agency, and it’s interim guidance, not American law. Nobody needs to wait for their own regulator to copy four controls that already work. And note where their own maturity ladder ends up for sensitive work. No external access, model hosted locally, on infrastructure you control. From dirt to desktop. Your data, your rules.

Laura: So the thread. A bar association asked for consent before the software listens. An examiner asked for the minutes. A cyber agency asked for an owner and a log. Two continents, three authorities, ten days, and none of them wanted a demo. Every one of them wanted proof. Capability is the part you buy. Evidence is the part you keep, and nobody can buy it for you afterward.

Arthur: So here’s our question this week. Pick the newest AI tool inside your business. Can you name the person accountable for it, and show what it did last week? If both come easily, you’re in better shape than most. If they don’t, that’s not a failure, it’s just the next thing to write down. Reply to the Modular Briefing email and tell us where you landed, or write to me at arthur at modtechgroup dot com. A person reads every reply.

Laura: Thanks for spending a few minutes with us.

Laura & Arthur: This has been The Modular Briefing. Your data, your rules. We will see you next time.

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. When we left you, we said Cale was headed to Ai4 in Las Vegas to see what actually holds up. He’s back. Three days, thirty booths, two legends on stage, and one sticker from the CIA. What he saw is the story the news wrote while he was there. When an agent acts, who answers for it? A federal court just gave an answer. So did a stress test of human oversight. So did a continent pricing its own dependency.

Arthur: Story one. On August fourth, the Ninth Circuit Court of Appeals answered the question every AI pilot eventually runs into. When an agent acts, who did it? The case was Amazon against Perplexity, whose Comet shopping agent had been operating on Amazon’s site on customers’ behalf. Amazon had won an order blocking it back in March. The appeals court threw that order out.

Laura: And the reasoning is the part your business should care about. Under the federal anti-hacking law, the court said access means a person entering a computer system. Software is a tool. Tools don’t carry liability. So when your employee points an agent at a system, the one who accessed it is your employee. And behind your employee, your firm. The honest caveats, because the legal press is being careful and so are we. One circuit, an early stage, and Amazon’s other legal theories are still open. This is not a court declaring agents fine. But it’s the first appellate word, and it points the accountability at whoever deployed the agent. So the vague question just became specific. Who at your firm may point an agent at what, and is that written down? If nobody wrote it down, then today the answer is everybody. Your data, your rules, and that goes for the AI itself. Your AI, your rules.

Arthur: Story two. Almost every AI policy written this year leans on one sentence. A human reviews it. This week The Register reported a number for how much weight that sentence holds. A Belgian developer, Alex Wauters, built a browser game that flashes real agent permission prompts, some safe, some dangerous, sixty seconds to approve or deny. Across forty thousand rounds, players approved roughly one dangerous request in three.

Laura: And before anyone says that’s just a game, Anthropic’s own telemetry says real users approve about ninety-three percent of what their agents ask for. Approval fatigue. The more prompts you see, the less you read each one. Cale heard the same conclusion from the defense side at Ai4. On the cybersecurity panel that stuck with him most, Ed Cartagena of Menlo Security laid out the new threat picture, and Cale’s takeaway was two sentences long. An agent with borrowed credentials behaves like a fast, tireless employee nobody supervises. And attacks move at machine speed, so a human clicking yes on every step was never going to be the control that holds. Your people aren’t careless. The control was never built to carry the whole load. So layer it. Least access, so a bad yes can’t reach client files. A boundary the agent can’t talk its way out of. Let the machines watch the machines, and save the humans for the calls a human should make.

Arthur: Story three, also from The Register. A survey of fifteen hundred businesses across the UK, France and Germany found nearly three in four are worried the US government could cut off their access to American technology. And more than half said that if they lost their cloud services, they could keep operating for one business day. One.

Laura: Fair disclosure, the way the reporting makes it. The survey was commissioned by Proton, a Swiss company that sells the sovereign alternative, so weigh it accordingly. But the gap is real, and it isn’t only European. Fewer than half of those firms have a continuity plan they’ve actually tested. So here’s the move. An exit plan is not a migration. It’s a document. Where does your data live? Who can reach it? What still runs if a vendor stops answering? How long would a move take? You can write the first version in an afternoon. And if you want the stronger position, run the work that matters on infrastructure you control, from dirt to desktop, so the kill switch question never has your name in it. Your data, your rules.

Arthur: Before we close, the report from the floor. Cale talked to every booth he could get to at Ai4, more than thirty. Infrastructure on one end, agent governance as far as he could see. Cisco. PayPal. Mistral. His read comes in three parts.

Laura: Part one, the argument won. A main stage keynote was literally titled From Renting AI to Owning Intelligence, appliance vendors are printing no tokens, no surprises on their booth walls, and the CIO of a NATO procurement agency described keeping local models close for the sensitive work. Owning your AI has quit being a niche position. Part two, a warning if you’re shopping that hall. Most of what’s sold as agent governance is a dashboard. It shows you what your agent did. Far fewer products will stop it, and watching an agent misbehave in real time is not a control. The crowd of new names reminded Cale of the late nineties, and his honest bet, take it as one, is that many won’t be around in a year. If your oversight lives in a startup’s dashboard, your controls inherit that startup’s odds. Part three, the keynotes. Andrew Ng and Geoffrey Hinton famously disagree about where this goes, and Cale will tell you Hinton’s doom ran a little thick. His take home fits in two sentences. Build now, the way Ng says. Govern like Hinton is right.

Arthur: So the thread. The court says an agent’s actions belong to whoever pointed it. The research says a tired human clicking yes isn’t ownership. Half a continent just learned that renting everything means owning nothing. And on that show floor, capability was everywhere and enforcement was scarce. Capability is easy to buy. Accountability has to be assigned.

Laura: Our question this week, and be honest. If one of your agents did something wrong yesterday, could you reconstruct what it touched, or would you be guessing? Reply to the Modular Briefing email and tell us which one you are. A person reads every reply.

Arthur: Thanks for spending a few minutes with us.

Laura & Arthur: This has been The Modular Briefing. Your data, your rules. We will see you next time.

Laura: 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 Laura.

Arthur: And I’m Arthur. Three stories today, and one question sitting underneath all of them. When the AI does something, whose name is on it? What your people are quietly using it for, what real oversight looks like, and what the law does when nobody is accountable.

Laura: Story one. This week a research team published an analysis of more than eight hundred thousand messages from people using a popular AI chatbot for work in the United States. Here is the number. Across every occupation they studied, forty-four percent of work messages were about tasks outside the sender’s own job. For people in human resources, it was sixty-nine percent. Third highest of any group.

Arthur: Sit with what that means. Roughly two out of three times an HR professional opens that tool at work, they are doing marketing, or engineering, or finance. Not HR. That is not people slacking. That is people solving a problem that landed on their desk with nobody to hand it to. The report found the crossover runs highest at small companies. Now the uncomfortable version. If your HR team is drafting finance work in a general public tool, what employee information went in with the prompt? A ban does not fix that, because the need is real and the need wins. A sanctioned workspace your company actually owns does, with one person accountable for what goes into it. Your data, your rules.

Laura: Story two. A pensions industry body published an AI governance framework this week for trustees and administrators. Five principles. Two of them are worth stealing no matter what business you are in.

Arthur: The first is proportionality. Instead of one blanket AI rule for the whole company, you sort uses into low, medium and high risk, and the high risk ones get real validation. Drafting a meeting summary is not the same as calculating somebody’s retirement benefit, and a policy that treats them identically gets ignored at both ends. The second is the one people skip. They call it meaningful human oversight, and the word doing the work is meaningful. The reviewer has to hold real authority to change the outcome. If the person reviewing cannot overturn the machine, you do not have oversight. You have a rubber stamp with a job title. That is the whole difference between a framework you can show an auditor and a framework you can actually run.

Laura: Story three, and it ties the week together. After an AI agent broke out of a test environment and went after another company’s systems, reporters put a simple question to the lawyers. Who is legally responsible when an AI agent attacks? The answer, so far, is that nobody knows.

Arthur: The reason is worth understanding. Our laws were built around human decision makers. They know how to ask about intent and oversight. An AI system is not a legal person, so it cannot carry any of that. One security strategist quoted in the piece laid out where the questions land instead. Who designed the system. Who decided what it was chasing. What safeguards were in place. How much independence somebody judged acceptable. Look at that list. Every one of those is answered by a person, or it does not get answered. The AI did it is not a defense. It is an unanswered question with your company’s name on it.

Laura: One more, and this one is from our own week.

Arthur: Cale wrote about a bank hiring its first Chief AI Officer, and the argument was the one we just made. Somebody has to own the decision. That post reached more than thirteen thousand people. He wrote the same argument two more times and posted it the next two mornings. Those reached about two hundred each.

Laura: Same person, same idea, same audience. So what changed?

Arthur: The day, and the first sentence. The one that traveled went out while the news was still news, and it opened with the bank, the name and the job title. Which is the same reason vague policies fail. Nobody can act on use AI responsibly.

Laura: That is the thread today. Your team is already using AI in ways nobody approved, real oversight means somebody can say no, and the law is going to come looking for a name. Have one ready.

Arthur: Quick note. Modular is at Ai4 in Las Vegas this week, August fourth through sixth. Cale is going for what is actually running in production, not what is announced from a stage. Whatever holds up, you will hear it here.

Arthur: And here is our question this week, and we want a real answer. At your shop, if somebody asked who approved the AI that touches your client files, is there a name? Or is there a document? Tell us. Reply to the Modular Briefing email and it lands with a person who reads it.

Laura: Thanks for spending a few minutes with us.

Laura & Arthur: This has been The Modular Briefing. Your data, your rules. We will see you next time.