AI Boosts Fraudsters’ Deception

Fraudsters are “practiced at the art of deception.”[1] And now AI has given them new tools to become faster, better and, yes, cheaper at it. They don’t even have to build AI tools themselves. According to reporting by Axios, they can purchase purpose-built packages on the dark web.
It has always been important for accounts payable to be alert to fraud. It’s only getting more crucial. But as any experienced manager can tell you, sustaining the necessary vigilance is difficult.
Crime tools, sold by subscription
The underground market for AI tools has exploded in less than a year. Research from Halcyon’s Ransomware Research Center, shared with Axios, counted fewer than 50 ads per month for AI tools on criminal forums and Telegram channels in late 2025. By February 2026 the count topped 1,400 a month.
The prices are startlingly low. Researchers found access to paid ChatGPT accounts offered for as little as 10 cents and prompts that trick mainstream AI models into ignoring their safety rules for 32 cents. Sellers now run storefronts and subscription plans that look a lot like legitimate software-as-a-service businesses.
Halcyon sorted what’s for sale into four categories:
- Jailbroken and stolen AI services — mainstream models with their guardrails stripped away.
- Identity fraud and phishing kits — tools that make it quick and easy to impersonate a real person.
- Models built for crime — such as WormGPT, designed to write phishing emails and malicious code.
- AI-powered infrastructure — including a system that can place scam calls to as many as 120 people at once.
Cynthia Kaiser, who leads the Halcyon research center and was formerly deputy director of FBI Cyber, told Axios the findings show “how cheap it is to conduct these activities.” Her point for finance teams: skills that once required a specialist can now be rented by almost anyone.
There is one reassuring note. Halcyon found no evidence of AI that can run an entire attack on its own. For now, AI is making the individual steps of familiar schemes cheaper, faster and more convincing.
That is enough to change the economics. An Interpol threat assessment, cited by The Tech Edvocate, found AI-enhanced fraud schemes produce roughly 4.5 times the profit of conventional ones. Scale, polish and speed all work in the criminal’s favor.
Where it hits accounts payable
Most AI-assisted fraud against AP isn’t new. It is the same old schemes, executed with fewer telltale flaws.
Fake and altered invoices. Fraudsters can now ask general-purpose chatbots to generate realistic receipt and invoice images, down to a convincingly crumpled look. Template websites and editing tools are also being used to create fake invoices or change the amounts on real PDFs. The typo-riddled, off-format invoice that used to give a fraud away is disappearing.
Bank account change requests. Requests from “suppliers” to update remittance details are a growing share of attempted fraud. Criminals register lookalike domains and slip into existing email threads. In some cases, a real supplier’s email system is compromised, so the fraudulent request comes from a legitimate address and every one of that supplier’s customers is exposed. AI writing tools make these messages read exactly like the vendor’s normal correspondence.
Executive impersonation. Deepfake audio and video can now reproduce a CFO’s or CEO’s face, voice and mannerisms. A rushed instruction to wire funds for a confidential deal is far harder to second-guess when it appears to come from the boss in person.
Synthetic vendors. AI can generate a plausible company name, a professional website, online reviews and even staff profiles with AI-generated headshots. A shell vendor that once would have failed a quick web search can now pass one.
What AP teams should do now
The strongest defenses are still process controls that a convincing fake can’t talk its way around.
- Assume fraud is already happening. Run historical payment and invoice data through an audit. Organizations that believe they have no fraud are often surprised by what turns up.
- Never verify through the channel the request came in on. Confirm any bank account change by calling the vendor at a number already on file, not one in the email. The same rule applies to an urgent payment instruction from an executive, however real the voice or video seems. (This is referred to as “out of band” verification.)
- Hold bank changes before the next payment. A waiting period plus a second approver for any remittance change defeats most account-takeover schemes.
- Verify new vendors beyond the website. Check tax ID matches, business registrations and references. A polished web presence is no longer evidence that a company exists.
- Use tools built for the job. General-purpose fraud filters may fall short. Look for solutions aimed at specific risks: invoice and document fraud, vendor bank-change fraud, or T&E abuse.
- Retrain staff on what “real” looks like now. Spelling errors and awkward phrasing are no longer reliable red flags. Teach the team to look for urgency, secrecy and any request to bypass normal procedure.
Beating the Art of Deception
AI has lowered the cost and increase the speed of deception and, unlike AI “art,” has also raised the quality of deception. Technology is working to give AP departments a way to sustain the vigilance that is hard to keep up “by hand.” The winners will be the teams that stay current on the deceiver’s methods, update and consistently follow controls, and utilize tools that provide vendor validations and include automatic out-of-band verifications.
To learn how VendorInfo can bolster your fraud controls, contact us.
[1] “You Can’t Always Get What You Want,” Mick Jagger and Keith Richards


