A familiar voice on the phone is no longer proof of anything. AI can clone a voice from a short public clip and speak live in it — to your bookkeeper, your office manager, or your spouse. This week's Scale with AI episode explains why you cannot out-listen a good fake, the one simple rule that stops these scams cold, and the November 25 opt-out deadline that lands in six days.

The Shift

For a hundred years, hearing a voice you knew was proof. Your manager called, your supplier called, and the sound itself settled the question. That proof is gone. Modern AI can learn the sound of a voice from a very short public clip — a video on your website, a podcast, a social post — and then speak any words in that voice, live, in a real conversation. It can also fake a face on a video call well enough to pass a quick glance. And you cannot train your ears out of this: a good fake does not sound robotic. It has the pauses, the accent, and the little phrases the real person uses. Telling your team to listen more carefully is not a plan. The businesses that are protected did not buy better ears. They changed the procedure, so that a voice, real or fake, is never enough on its own to move money or change records. That change in thinking — from trusting the sound to trusting the process — is the whole shift.

The Law

Two clocks are running. First, the Federal Communications Commission's consent revocation rules take effect November 25 — six days from now. If your business calls or texts customers, a customer who says stop, quit, end, revoke, opt out, cancel, or unsubscribe must be honored, and honored within ten business days at the very latest. The safer practice is to honor it the same day, log it, and keep the proof. If you use an exclusive opt-out method, it has to be disclosed clearly on every call and every text — so test yours this week, before the deadline lands in the middle of your busiest season. Second, a wider warning, labeled honestly: this week California proposed mandatory safety standards and emergency shutdown powers for the largest AI developers. That is a proposal, not a law, and it aims at the labs, not at your shop. The reason it matters is direction: regulators everywhere are moving from trusting companies to verifying them. The businesses that can show a written process — who approved what, and how a customer opted out — will be the ones who sleep well. This is general information, not legal advice.

Who's Doing It

The FBI's internet crime report for 2025 recorded 24,768 business email compromise complaints, with reported losses of just over $3 billion. Its first-ever AI section counted reported AI-related losses of about $893 million across fraud types. Independent tracking put deepfake-related fraud losses at about $1.1 billion globally in 2025, nearly triple the year before — and those are reported figures; experts agree most incidents are never reported at all. Against numbers like that, look at what the businesses that caught these attacks actually did. The pattern in case after case is not a detection tool. It is a callback. A bookkeeper gets an urgent voice message that sounds exactly like the owner, asking for a payment to a new account. Instead of paying, she hangs up and calls the owner back on the number she already had saved long before that message arrived. The cloned voice can pass a listening test. It cannot answer a phone it does not control. Owners are also adding a simple spoken code word for money requests — a word that is never written in email — and a two-person rule, where no single person can both approve and send a payment. None of that requires new software. It requires one decision, made before the scary call comes, not during it.

Your Move

Three steps, all doable this week. Step one: adopt the Known Number Rule today. Any request to move money, change payment details, or share a code gets confirmed by calling a number you already held before the request arrived — never a number supplied inside the request. Step two: pick your code word and your two-person rule, and tell your team in writing. Keep it to one page, in plain words, where everyone can find it. Step three: test your opt-out before November 25. Send the word stop to your own business number and watch what happens. If it does not work cleanly, fix it now, while it is a drill and not a complaint. We walk through the whole picture in today's episode of Scale with AI: listen to the full episode.

That is exactly the work we do at The Agency. We build the system, set the approval rules, keep you inside the law, and put a human approval gate on every promise that matters. You do not need to learn these tools — that is our job. Start with a free AI Audit, and we will show you exactly where your business is exposed and what to fix first.