FraudGPT: the $200-a-month "criminal ChatGPT" that was probably a scam itself (2023)

FraudGPT was sold in mid-2023 as "ChatGPT with no boundaries" for criminals: $200 a month or $1,700 a year for a bot that would write phishing emails, build fake login pages, churn out malware and find cardable websites. A seller using the alias CanadianKingpin12 pushed it on a Telegram channel created on 23 June 2023 and across half a dozen dark-web markets, claiming more than 3,000 sales. Security firm Netenrich documented the listings on 25 July 2023 and the name went everywhere. Then in 2025 Cisco Talos tried to actually buy it. The login details they were given didn't work. The seller asked for more crypto to pay for a "crack". Talos concluded there was probably no product at all. Status: Seller unidentified; no confirmed attack ever tied to FraudGPT output; assessed by Talos as a scam on criminals.

The pitch

Netenrich framed FraudGPT as a sibling of WormGPT, which SlashNext had reported on 13 July 2023, and floated a possible link between the two operators while being clear that was speculation. Netenrich also said it knew of no confirmed attack using FraudGPT content. The day after its report, FBI Director Christopher Wray warned publicly about criminals using generative AI for malware and spear phishing, without naming any product.

Tip: a feature list is not a demo. If nobody outside the seller has shown the tool producing anything, treat the capability as unproven.

The purchase test

Two years later, Cisco Talos did what nobody else had: tried to buy it. After a long negotiation on Telegram, CanadianKingpin12 handed over a username and password for the FraudGPT site. They didn't work. The seller's answer was to ask for cryptocurrency to buy a "crack" for the login page.

That's an advance-fee scam with the goalposts moved. Talos found other would-be buyers on the same forums reporting the same experience and assessed that the actor had no working product and was simply taking money from other criminals under the FraudGPT name. Outpost24 separately noted a spread of fake Telegram channels promising tools that never arrived.

One caveat. Rapid7's June 2026 report on the criminal AI market gives a different account of FraudGPT's capabilities. It doesn't, though, independently establish that the original 2023 offering worked. The honest summary is that the strongest hands-on evidence says scam, and nobody has produced evidence to the contrary.

Why it matters

FraudGPT sits in almost every threat-intel slide deck from 2023 as proof that AI was lowering the bar for phishing. It may be better evidence of something else: how quickly an unverified underground advert becomes an industry fact.

There are two social-engineering layers here. The advertised one, AI making lures more convincing at scale, is real as a category even if this product wasn't. The demonstrated one is that the market for criminal AI tools runs on the same urgency, scarcity and fake social proof as any scam aimed at the public. The marks were just other criminals.

What stops it

• Be sceptical of "blackhat AI" claims, whether FraudGPT, WormGPT, GhostGPT or the next one. Ask for independent testing before repeating a capability.
• Defend the delivery mechanism, not the branding. A phishing email is a phishing email whoever or whatever wrote it.
• DMARC, SPF and DKIM enforcement, plus link and attachment sandboxing at click time.
• Out-of-band verification for any payment or bank-detail change. Still the control that beats BEC.
• Train on urgency plus money, not on spotting "AI writing". Polished AI text and polished human text look the same.
• Tune detection to BEC behaviour and infrastructure rather than hunting for an AI signature that doesn't exist.

See also AI-generated malware, agentic AI attacks and phishing.

Common questions