TL;DR: AI now lets criminals clone an executive’s voice and write flawless fake invoices that blend into normal accounts payable workflows. Business email compromise cost US businesses over $3 billion in 2025. You can’t train your way out of fraud you can’t see, so the real fix is process: out-of-band verification, layered access controls, and a culture that rewards slowing down on payments.
Picture a finance employee on a routine video call with the CFO and a few other executives. The request is urgent but normal, so the money goes out. Every person on that call was a deepfake. That’s not a hypothetical. It’s roughly how attackers stole $25.6 million from engineering firm Arup, and it shows where business email compromise is heading.
The numbers are sobering. According to the FBI’s 2025 Internet Crime Report, business email compromise (BEC) cost US businesses more than $3 billion in a single year, making it one of the most financially damaging cybercrimes on record. The same report logged AI-enabled scams for the first time, with over 22,000 complaints and more than $893 million in losses.
AI has made these attacks much harder to spot. The question for your accounts payable team is no longer whether they can catch a suspicious request. It’s whether your payment process makes fraud hard to pull off, no matter how convincing it looks.
Why Are Accounts Payable Teams a Top Target?
Accounts payable is a top target because it sits at the intersection of trust and time pressure. AP teams handle invoices, update supplier details, and send payments, often while juggling high volume and tight deadlines. For an attacker, that mix is close to perfect.
Most successful fraud doesn’t involve breaking into your systems. It relies on impersonation. The attacker poses as a trusted executive, supplier, or coworker, then redirects a payment or changes bank details before anyone notices.
AI has made that impersonation far easier to scale. Crafting a convincing request used to take skill and time. Now widely available tools handle the research, the writing, and the small details that help fraud blend into normal work. A weakness in your email and spam protection only widens the opening.
What Does AI-Enhanced Invoice Fraud Look Like?
AI-enhanced fraud shows up in three main ways: emails that match your normal workflow, payment redirection, and voice or video cloning. Each one is built to feel routine, which is exactly why it works.
Emails that blend in. Old phishing leaned on volume and sloppy writing. AI fixed both. Modern BEC emails are grammatically clean and written in the tone of the person being impersonated. They reference real projects, current invoice numbers, and upcoming payment runs. By mid-2024, an estimated 40% of BEC phishing emails were already AI-generated, and that share keeps climbing.
Payment redirection. This is one of the most common patterns. Attackers quietly change the destination account on a real invoice, or send a short note claiming a supplier updated its banking details. The surrounding content looks legitimate because it’s often pulled from real correspondence. Many of these attacks start with account takeover, where a criminal slips into a real inbox and watches before striking.
Voice and video cloning. Email isn’t the only channel anymore. AI can clone a person’s voice from seconds of audio, which makes a fake voicemail or live call sound like a known executive. For teams used to approving urgent payments by phone, that erases one of the few checks email security can’t cover.
Why Don’t Traditional Security Checks Catch This Anymore?
Traditional checks no longer work because the warning signs they relied on are gone. Awkward phrasing, mismatched logos, odd sender addresses, and generic greetings were the tells. AI has removed all of them.
Today’s fraud emails can name your suppliers, reference real invoice values, and match your company’s voice. When a fake request is impossible to tell apart from a real one, asking your AP team to spot it is asking the wrong thing of the wrong people.
The businesses that cut their risk aren’t telling staff to be more suspicious. They’re building verification that works no matter how a message looks. Strong security awareness training still has a place, but it can’t be your only line of defense.
How Do You Actually Stop AI Invoice Fraud?
You stop AI invoice fraud by removing ambiguity from high-risk actions. The goal isn’t sharper instincts. It’s a payment process that makes fraud difficult by design.
Make out-of-band verification standard. Any request to change bank details or approve an urgent, off-cycle payment should require a second confirmation through a known, independent channel. Not a reply to the same email thread. Call the supplier on a number already on file, or confirm with the person directly. That one step breaks the impersonation chain, and it costs nothing but a written procedure and the habit of following it.
Layer your access and authentication controls. Limiting who can touch financial systems and enforcing multi-factor authentication contains the damage a single compromised account can do. If an attacker gets into a vendor’s email, controls on your end add friction that can stop a fraudulent change before money moves. This is where layered cybersecurity protection earns its keep.
Build a culture that supports slowing down. Fraud prevention improves when staff feel safe questioning a request, even one that appears to come from the top. An employee who pauses a payment to verify it isn’t being difficult. They’re doing exactly what good process requires. That culture starts with leadership modeling it and making clear that slowing down on high-risk payments is always the right call. Setting up a clear verification process gives your team permission to act on it.
The Real Fix: Shift the Burden From People to Process
A few things are worth holding onto. AI fraud succeeds because it looks normal, not because your people are careless. The damage almost always runs through a payment, so that’s where your controls belong. And the tools attackers use are advancing fast, but the controls that stop them don’t have to be complicated. They have to be consistent.
You don’t have to build this alone. As part of our managed IT services in Louisville, we help finance and leadership teams put these controls in place before an attacker tests them for you. To review your current payment and verification process, schedule an intro call with our team.
Frequently Asked Questions
What is business email compromise (BEC)?
Business email compromise is a scam where an attacker impersonates a trusted executive, supplier, or coworker to trick someone into sending money or changing payment details. It usually relies on social engineering rather than malware or malicious links. BEC was one of the most costly cybercrimes of 2025, with US losses topping $3 billion.
How is AI making invoice and payment fraud worse?
AI lets attackers write flawless, personalized emails and clone voices at scale. Requests now reference real projects, suppliers, and invoice numbers, so they blend into normal work. The old warning signs like typos and odd phrasing are largely gone, which makes fraud far harder to catch by eye.
Can deepfake voice calls really fool my finance team?
Yes. AI can clone a convincing voice from just seconds of public audio, like a webinar or podcast clip. That makes fake voicemails and live calls sound like a real executive. In one case, criminals used deepfakes on a video call to steal $25.6 million, which is why phone and video approvals need a backup check.
What is out-of-band verification and why does it matter?
Out-of-band verification means confirming a request through a different, trusted channel than the one it came in on. If you get an email asking to change bank details, you call the supplier on a number you already have on file. It matters because it breaks the impersonation chain, no matter how real the original request looks.
Does security awareness training still help against AI fraud?
Yes, but it can’t be your only defense. Training builds good habits and helps staff recognize pressure tactics. Since AI fraud can be impossible to spot by sight alone, training works best paired with process controls. Our cybersecurity essentials guide covers how the layers fit together.
Let’s Pressure-Test Your Payment Process
You shouldn’t have to wonder whether one convincing email or phone call could move money out of your accounts. If you’re ready to find the gaps in your verification process before an attacker does, we’re glad to walk through it with you. Reach out to the Z-JAK team here.