How do AI-powered scams work and how to prevent it

It’s May 2026, and we’re halfway through Fraud Prevention Month here in Canada. This year’s theme, “The Hidden Game,” lands harder than any campaign before it. The old tells are gone. No more broken English. No more grainy stock photos. No more robotic voices on the other end of the line. AI has stripped fraud of every warning sign we used to teach employees to spot.

The numbers from KPMG Canada’s latest report make the stakes clear. 81% of Canadian businesses that experienced fraud this past year faced an AI-enabled attack. 72% of companies lost up to 5% of annual profits to these hidden crimes. That isn’t a security problem anymore. That’s a balance sheet problem.

Here’s how the 2026 playbook works, and how Novus is helping Canadian businesses turn AI into the defender instead of the attacker.

What AI-Powered Scams Actually Look Like Now

Scammers stopped spraying and praying years ago. Today they run industrialized AI toolchains built for precision.

Voice cloning. Three seconds of audio. That’s the threshold. A clip from a podcast, a LinkedIn video, a conference keynote, and an attacker can clone an executive’s voice with 95% accuracy. Then they call your AP team to authorize an urgent wire transfer, and the voice on the line sounds exactly like the CEO.

Deepfake documents. AI generates fraudulent invoices and contracts that mirror your company’s branding, internal terminology, vendor codes, and signatures. Everything looks “all green” in your review workflow because it was designed to.

Synthetic identities. Fraudsters stitch together real stolen data with AI-generated social media footprints, credit histories, and reference networks. These Frankenstein profiles regularly clear standard bank onboarding because every individual data point checks out.

Emotional bot-nurturing. LLMs let one scammer maintain dozens of long-running, emotionally intelligent conversations at once. Romance scams, fake recruiter pipelines, investment grooming. The bot adapts its tone in real time, builds trust over weeks, and only asks for money after the relationship feels real.

It’s Also a Customer Experience Problem

Before we get to the defense playbook, one thing worth naming. Fraud in 2026 isn’t only a security problem. It’s a customer experience problem, because every synthetic account, fake review, and bot conversation chips away at the trust real customers have in your brand. The companies treating fraud as a CX issue, not just an IT issue, are the ones holding onto loyalty while their competitors leak it.

Fighting Fire With Fire: How AI Beats AI

Here’s the good news. 52% of Canadian organizations are already fighting AI with AI, and the defensive side of the technology is moving just as fast as the offensive side. At Novus, we build these layers directly into client workflows.

Behavioral fingerprinting. AI isn’t just watching what a user does. It’s watching how they do it. We integrate models that analyze typing cadence, mouse movement, and navigation patterns. The moment an “authenticated” user starts behaving like a bot or an impostor, the system triggers step-up authentication automatically.

Forensic liveness detection. Humans can no longer reliably tell a deepfake from a real person. So we stop asking them to. Our verification flows use AI to scan for biological signals like blood flow patterns in a video feed and microscopic acoustic inconsistencies in a voice call. These are the artifacts that today’s clones still can’t fake.

Agentic security agents. Passive firewalls aren’t enough anymore. We deploy active defense agents that interrogate the environment in real time. They scan incoming communications for scam language patterns, urgency tactics, and gift card requests, then warn employees with on-screen alerts before a single link gets clicked. The same approach is now being applied to customer-facing funnels to filter out synthetic traffic and bot fraud at the front door.

Three Things You Can Do This Week

You don’t have to overhaul your stack to cut your risk meaningfully. Start here.

  1. Set a business code word. For any unusual voice or video request involving money, require a pre-shared, offline code word to verify identity. Simple, free, and devastating to a voice clone.
  2. Move to passkeys. Passwords are easy for AI to phish or crack. Passkeys use cryptographic keys stored on the device, which makes deepfake impersonation nearly useless at the login layer.
  3. Require dual approval. No high-value transaction should ever be triggered by a single person on a single channel. One email, one call, one DM is never enough.

The 2022 Defenses Are Gone

If your business is still running fraud defenses built for a 2022 threat model, you’re not just behind. You’re a target. The attackers have already upgraded. The question is whether your workflows have.

Novus builds secure, AI-powered operations for businesses across O&G, FinTech, healthcare, and AgriTech. If you want to stop being part of the “hidden crime” statistics, let’s talk.

author avatar
Ysa-Novus