How do AI scams work? Criminals use generative AI to write convincing phishing emails, clone a real person's voice from a short audio clip, and impersonate executives on video calls. The goal is almost always the same: get someone to send money, change banking details, or hand over a password. The strongest defence is not spotting the fake. It is a verification process that still works when the fake is perfect.
For years, the advice for spotting a scam was simple: look for spelling mistakes, strange greetings, and requests that do not sound like the person they claim to come from. That advice is now dangerously out of date. AI tools can write flawless email in any tone, imitate the way your supplier writes, and reproduce your business partner's voice well enough to fool the people who know them best.
Small businesses are an attractive target. Money-moving authority is usually concentrated in one or two people, approvals happen over text and phone, and there is rarely a security team watching for anything unusual. The good news is that the fixes are cheap, simple, and mostly about process rather than technology.
The numbers are sobering. Fraud reported to the Canadian Anti-Fraud Centre hit a record of more than $704 million in losses in 2025, and authorities estimate that only 5 to 10 per cent of fraud is ever reported. Spear phishing, which includes criminals impersonating an executive or supplier to redirect payments, was the second-costliest type of fraud that year at $67.9 million. And when a business is hit, it hurts: in 2024, the average reported loss per business incident was about $110,000.
What AI Has Actually Changed
The scams themselves are not new. Fake invoices, "urgent" requests from the boss, and bogus tech support calls have been around for decades. What AI has changed is how convincing and how cheap they are to run. Canada's National Cyber Threat Assessment warns that AI is lowering the barrier to entry for cybercriminals and making their social engineering more convincing.
Phishing Without the Tells
Generative AI writes clean, natural, well-formatted email in English or French, tailored to your industry and even to specific people in your company. An attacker can research your business from your website and LinkedIn, then produce a message that references a real project, a real supplier, and a real colleague, in seconds and at scale.
Voice Cloning
Modern voice-cloning tools can produce a convincing imitation of someone from a short sample of clear audio. A voicemail greeting, a social media video, a podcast interview, or a recorded webinar is often enough. The result is a phone call that sounds exactly like the owner asking the bookkeeper to "get this payment out today."
Deepfake Video Calls
The most widely reported case involved the engineering firm Arup. In early 2024, an employee in its Hong Kong office joined a video call with what appeared to be the company's chief financial officer and several colleagues. Every other person on the call was a deepfake. The employee made 15 transfers totalling about HK$200 million (roughly US$25.6 million) before the fraud was discovered. If a multi-person video call can be faked, a single phone call certainly can.
Faster, Better-Researched Pretexts
AI also speeds up the homework. Attackers use it to summarize who works where, who approves payments, which suppliers you use, and when key people are travelling (often from your own social media). The more a message fits your real situation, the less likely anyone is to question it.
The Scams Hitting Small Businesses Most
| Scam | How AI Makes It Worse | What Stops It |
|---|---|---|
| Supplier banking change | Perfectly written "we've changed banks" email, sometimes from a hijacked real supplier account | Call back on a number already on file before changing any payee details |
| Urgent request from the owner | Cloned voice or convincing text asking for a wire, e-transfer, or gift cards | A verbal passphrase and a rule that urgency never skips approval |
| Payroll diversion | An "employee" emails to update their direct-deposit account | Verify in person or by a known phone number, never by reply |
| Fake login pages | Pixel-perfect Microsoft 365 or bank sign-in pages sent by email or text | Phishing-resistant MFA such as passkeys or security keys |
| Fake tech support | A realistic caller "from Microsoft" or "from your IT provider" asking for remote access | Only allow remote access that you initiated, through a provider you already use |
Why "Spot the Deepfake" Training Is Not Enough
It is tempting to train staff to listen for robotic voices or watch for glitchy video. That helps a little today, but the technology improves every few months, and people under time pressure are poor detectors. A process that depends on someone noticing a fake will eventually fail.
The Canadian Centre for Cyber Security puts it plainly: "Audio or video alone should not be considered proof of identity." Its advice is to confirm sensitive requests through a second, independent channel.
The better approach is to assume the fake will be perfect and design your payment and access processes so that no single email, text, or phone call can move money or change banking details on its own. Verification beats detection.
A Seven-Step Protection Plan for Small Businesses
- Adopt a call-back rule. Any new payee, change to banking details, or unusual payment request gets confirmed by phone using a number you already have on file. Never use contact details from the message making the request.
- Require two people for payments above a threshold. Pick a dollar amount that fits your business. Anything above it, and every first payment to a new payee, needs a second person to approve.
- Set up a verbal passphrase. The owner and whoever handles payments agree on a phrase that is never written in email or chat. If "the owner" calls asking for an urgent transfer and cannot give the phrase, the answer is no.
- Use phishing-resistant MFA. Turn on multi-factor authentication for email, banking, accounting, and payroll. Where you can, use passkeys, security keys, or app prompts with number matching instead of text-message codes.
- Lock down your email domain. Publish SPF, DKIM, and DMARC records so criminals cannot easily send email that appears to come from your domain. This protects your customers as much as it protects you.
- Train with real examples, and make it safe to slow down. Show staff what current AI-written phishing looks like. More importantly, make it clear that nobody gets in trouble for verifying a request, even if it really was the boss.
- Limit what attackers can learn. You do not need to disappear from the internet, but think twice about publicly posting who approves payments, detailed org charts, or real-time travel plans.
If It Happens: Your First-Hour Checklist
Speed matters. The sooner a fraudulent payment is reported, the better the chance your bank can stop or recall it.
- Call your bank immediately and ask them to stop or recall the payment. Do this before anything else.
- Secure the accounts involved. Reset passwords, sign out all active sessions, and check your email for forwarding rules you did not create. Attackers often add these to hide replies.
- Preserve the evidence. Keep the emails, texts, call logs, and payment records. Do not delete anything.
- Report it to your local police and to the Canadian Anti-Fraud Centre (1-888-495-8501 or online).
- Check whether personal information was exposed. If customer or employee information was compromised, you may have breach-reporting and notification obligations under Canada's private-sector privacy law (PIPEDA).
- Call your insurance broker. Some cyber policies exclude or cap losses from "social engineering" fraud, so ask specifically about funds-transfer fraud coverage.
Fraud in Canada is badly under-reported, largely because victims are embarrassed. Reporting helps police connect cases and warn other businesses, and it is often required for insurance claims.
Frequently Asked Questions
Can AI really clone someone's voice from a short recording?
Yes. Modern voice-cloning tools can produce a convincing imitation from a short sample of clear audio, such as a voicemail greeting, a social media video, or a recorded webinar. That is why a familiar-sounding voice on the phone should never be treated as proof of identity for payments or banking changes.
How should we verify a request to change a supplier's banking details?
Call the supplier using a phone number you already have on file, never one provided in the email or letter requesting the change. Confirm with a person you know, have a second person approve the change, and consider a small test payment before sending a large amount to new banking details.
Where should an Ontario business report AI-enabled fraud?
Contact your bank first to try to stop or recall the payment. Then report the fraud to your local police and to the Canadian Anti-Fraud Centre at 1-888-495-8501 or through its online reporting system. If personal information was compromised, assess your breach-reporting obligations under PIPEDA.
Related Articles
Want a Second Set of Eyes on Your Defences?
ZABLEY sets up phishing-resistant MFA, email authentication (SPF, DKIM and DMARC), and simple payment-verification processes for small businesses across Ontario. We will show you where you are exposed and fix it in plain language.
Key Takeaways
- AI has removed the classic warning signs of scams: phishing emails are now well written and voices can be cloned from short clips
- Reported fraud losses in Canada hit a record of more than $704 million in 2025, and most fraud is never reported
- Deepfake video calls are real: an Arup employee transferred about US$25.6 million after a call with deepfaked colleagues
- Design processes that assume the fake is perfect: verification beats detection
- Use a call-back rule, two-person approval above a set amount, and a verbal passphrase for urgent requests
- Phishing-resistant MFA and SPF, DKIM and DMARC on your email domain block the most common entry points
- If fraud happens, call your bank first, then secure accounts, preserve evidence, and report to police and the Canadian Anti-Fraud Centre