AI scams aimed at small businesses, and the call-back rule that stops them

The short version

Scammers now use AI to write convincing emails and clone voices. The defense is old and cheap: never move money on a message alone. Here's the rule, and the FBI numbers behind it.

Picture this. Your bookkeeper gets an email from "you" asking her to wire a deposit to a supplier today. The writing sounds like you. The signature is right. Maybe there's even a quick voicemail in what sounds like your voice.

It isn't you. That's the core of what the FBI calls business email compromise, and AI is making it easier to pull off. Writing a smooth, specific, error-free email used to take a scammer effort. Now it takes a prompt.

You can't spot these by reading closely anymore. You stop them with a process. This post covers what's happening, the FBI numbers, the scams to expect, and one rule that handles most of them.

What the FBI's numbers say

The FBI's Internet Crime Complaint Center (IC3) 2025 annual report is the best public count we have. It only covers what people reported, so real losses are likely higher. Here's what it says:

  • Overall: 1,008,597 complaints and $20.877 billion in reported losses in 2025, a 26% rise in losses from 2024.
  • Business email compromise (BEC): 24,768 complaints and $3,046,598,558 in losses. In 2024 it was 21,442 complaints and $2,770,151,146.
  • AI-related: more than 22,000 complaints that mention AI, with adjusted losses above $893 million. That's across all fraud types, and the biggest dollar share is investment scams, not business fraud.
  • BEC with AI: businesses reported over $30 million in losses to BEC scams involving AI in 2025. The report says not all BEC tactics are AI-enabled.

The FBI's definition of BEC: a scam targeting businesses or individuals who work with suppliers or regularly make wire transfers. Fraudsters get into email accounts and other channels, including phone numbers and virtual meeting apps, and use social engineering to move money.

So AI isn't most of the problem yet. BEC is a $3 billion a year problem with or without AI. What AI changes is how believable the message is. The report also says AI can write official-sounding emails that mimic a CEO, and that voice cloning can be used to request a wire payment.

The scams to expect

The fake boss or owner

A message that looks like it's from you, or a manager, asks someone to pay something or buy gift cards. It may be an email, a text, or a call with a cloned voice. It's usually urgent and a little secret: "I'm in a meeting, handle this quietly."

The fake invoice

A supplier or contractor you actually use "changes their bank details." The email thread looks real because the attacker may be inside the supplier's mailbox. The invoice is right. The account number isn't.

The cloned voice

AI can copy a voice from a short clip. Your voicemail greeting, a video on your site, a social post. A caller who sounds like you, or like a vendor, asks for a payment or a password. The FBI report describes voice cloning used to request wire payments.

The email takeover

Someone gets into a real inbox and watches. They wait for a real invoice, then reply to the real thread with new payment details.

The fake candidate or fake vendor

The FBI report describes people using voice spoofing, and possibly deepfakes, in online job interviews. In those cases it said the goal appeared to be getting into private computer networks. If you hire remotely, be aware of it.

The "your account has a problem" text

Not new, but AI helps scammers write it in clean English and match your bank's tone. Don't tap the link. Go to the site yourself.

The call-back rule

This is the one we'd put in writing in every business.

No money moves, and no account details change, on the strength of a message alone. Before you act, call the person back on a number you already had.

That's it. A few specifics:

  1. Use a number from your own records. Not the number in the email. Not the number that just called you. Your contact list, the last invoice you paid, the supplier's website you typed yourself.
  2. Any change to bank or payment details gets a call-back, even from a vendor you've used for ten years. Especially then.
  3. Two people approve wires and large payments. The person who enters it isn't the person who approves it.
  4. Urgent and secret is a red flag. Anyone who says "don't tell anyone" or "do it right now" is describing a scam. A real boss will wait five minutes for you to call back.
  5. A family code word helps too, if a "relative in trouble" ever calls. The FBI report notes voice cloning is used in distress scams mimicking a loved one, with losses above $5 million in 2025 for that type.

Why it works: the scammer controls the email, the text and maybe the voice. They don't control the phone number you've had for years. A call-back breaks the one thing they rely on, which is you answering inside their channel.

It also doesn't depend on spotting an AI fake, which is getting harder. The FBI report itself says AI-enabled content is becoming harder to detect and easier to make.

Set it up in an afternoon

  • Write the rule on one page and give it to everyone who touches money, email or customer accounts. That includes the person who answers the phone.
  • Turn on two-step sign-in for email, banking, your accounting tool and your domain account.
  • Keep a short list of verified numbers for your top vendors and your bank, kept offline or in a password manager.
  • Limit who can change payment details in your accounting software.
  • Practice once. Send a fake "urgent wire" test to your bookkeeper. See if she calls back. Don't make it a gotcha; make it a drill.
  • Read your insurance policy. Some policies limit or exclude losses from social engineering and wire fraud unless you've added coverage. We can't read your policy for you, so ask your agent directly.
  • Watch what you publish. A long video of you talking is voice sample material. You don't have to hide it. You do have to have the call-back rule.

Your staff's AI habits matter too. If they're pasting customer information into free tools, that's a leak of a different kind. See what your staff is pasting into ChatGPT and is ChatGPT safe for business data.

If you do get hit, call your bank right away and ask about a recall, then report it to the FBI at ic3.gov. Speed matters, so call the bank before you finish reading anything else.

If you also text or call customers, the consent rules are their own topic: see texting rules for contractors.

Frequently asked questions

Are AI scams actually common for small businesses?

The FBI's 2025 report counts 24,768 business email compromise complaints and over $3 billion in losses. It also counts over 22,000 complaints with AI references across all fraud types. Most BEC isn't AI-enabled yet, but AI makes messages more convincing. See how to protect business data with AI tools.

Can someone really clone my voice?

The FBI report says voice cloning can be used to request wire payments and to impersonate a loved one in distress. It describes how, but doesn't tell you how easy it is for any one person. Assume a clip of your voice online is enough, and rely on call-backs, not on how a voice sounds.

What is the call-back rule?

Never move money or change payment details because of a message alone. Call the person back on a number you already had, not one from the message. Add a second approver for wires. It works whether the fake is an email, text or cloned voice.

What do I do if I already sent money?

Call your bank right away and ask about a recall. Then file a report at ic3.gov. Speed matters most. Also change passwords on any account you think was exposed. This isn't legal advice, and your bank can tell you what options exist for your account.

How Di-Hy fits

We're not a security firm, and we won't promise you're unhackable. What we do is set up your tools sensibly and train your team on habits like the call-back rule and what not to paste into free AI tools. Training your team is step 3 of how we work with you, and Keep It Running means someone checks access and settings every month. We'll also tell you if a task shouldn't be handed to AI. If you want a one-page AI use rule for your staff, the free AI use policy generator is a start.

Next step: write the call-back rule on one page today. If you want help rolling it out, book a call.

This is general information, not security or legal advice. FBI figures are reported complaints and understate real losses. Read in November 2026.

Charles Holmes IV
Written by Charles Holmes IV
Founder of Di-Hy Digital, based in Ybor City, Tampa. He sets up and runs the systems he writes about.
More about us →

Want us to take a look?

Tell us what's not working. Charles reads every message and replies himself. The first call is free.

Thanks, it's in.

We'll be in touch soon. Prefer email? dihydigital@gmail.com

Want us to look at yours?

Book a free 30-minute call. We'll look at how customers find you, how fast they hear back, and what's worth fixing first.

Book a Free Strategy Call →