Phishing Deepfake & Synthetic Media Corroborated research/advisory

FBI PSA241203: how criminals use generative AI in financial fraud (2024)

The FBI's December 2024 warning on AI-enabled fraud: AI text, fake IDs, cloned voices and deepfake video, and what the FBI says to do about each.

Social Engineering Examples·4 sources

On 3 December 2024 the FBI's Internet Crime Complaint Center published PSA241203, a public warning titled "Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud". It isn't about one incident. It pulls together what the FBI was seeing across thousands of victim complaints: scammers using AI-written text, AI-made photos and ID cards, cloned voices and deepfake video to run the same old frauds faster and more convincingly. The advisory gives no loss figure. A year later, the 2025 IC3 annual report counted 22,364 complaints tagged as AI-related, with $893 million in adjusted losses. Status: an advisory, so no named victims and nobody charged.

Key takeaways

The FBI's core point: AI removes the tells. Bad grammar and inconsistent stories used to give scams away. Not anymore.
Text, pictures, voice, video. Each gets bolted onto a fraud that already existed, from romance scams to fake bank calls.
The fix the FBI leads with is almost quaint. Agree a secret word with your family.
The $893 million figure is from the 2025 annual report, not the advisory. Don't attribute it to this PSA.

The four ways AI shows up in fraud

Modality	Examples the FBI lists
Text	Phishing scripts, fake social media profiles, scam messages translated into fluent English, content for bogus investment sites, chatbots that keep victims talking
Images	Profile photos for fake identities, forged IDs and law-enforcement or bank credentials, fake celebrity endorsements, invented disaster photos for charity scams, sextortion images
Audio	Short cloned clips of a relative's voice used in staged crisis calls, and cloned voices used to impersonate account holders to their bank
Video	Real-time deepfake calls posing as executives, police or other authority figures; video "proof" that an online contact is real; deepfakes in investment promos

None of this is a new crime. Romance scams, grandparent scams, CEO fraud, dodgy investments: all older than ChatGPT. What's changed is the friction. Less work per victim. More victims per scammer. And far fewer slip-ups for a target to spot.

Tip: if a message, photo or call feels unusually polished for the person it claims to come from, treat that polish as a warning, not reassurance.

Voice and video in particular

Two of the FBI's examples deserve a closer look because they turn up again and again in the cases on this site. The first is the crisis call. A scammer clones a loved one's voice from a few seconds of public audio and rings a relative claiming to be in trouble and needing money now. See the Quebec grandparent case for what that looks like in practice.

The second is the bank impersonation, in reverse: a cloned voice used to pose as the customer when calling the bank, to get into the account. The PSA doesn't name a product or describe how a voice-authentication system might be beaten, and this page doesn't either.

On video, the advisory points to live deepfake calls impersonating executives. The Arup case earlier in 2024 is the obvious reference point.

The numbers, and what they aren't

Figure (2025 IC3 annual report)	Amount
Complaints tagged "AI related"	22,364
Adjusted losses, all AI-related	$893,346,472
Of which: investment fraud	$632,041,188
Of which: business email compromise	$30,256,592
Of which: confidence/romance fraud	$19,041,653
Of which: employment fraud	$12,550,185

Three caveats. These come from the 2025 annual report, a year after the PSA, which itself gives no numbers. "AI related" is a descriptor attached to a complaint that already has a crime type, so these dollars sit inside the annual total rather than on top of it. And complaint data isn't measured loss. It's what victims reported. For the full breakdown, see the IC3 2025 AI fraud page.

What the FBI says to do

Agree a secret word or phrase with family so a crisis call can be checked in seconds.
Hang up and call back on a number you already had. Never the one that rang you.
Does it sound like them? Odd word choices or a flat tone are worth a pause.
Check the picture. Hands with too many fingers, teeth that don't line up, shadows going the wrong way, video that lags.
Limit your public voice and image. Private accounts and restricted followers mean less raw material for a clone.
Never send money, gift cards or crypto to someone you've only met online or by phone.
Report to IC3 with full details. Aggregate complaints are how advisories like this one get written.

Common questions

What is FBI PSA241203?

A public service announcement from the FBI's IC3, published 3 December 2024, describing how criminals use generative AI text, images, voice cloning and deepfake video to make financial fraud more scalable and believable.

How much money has been lost to AI-enabled fraud?

The PSA gives no figure. The FBI's 2025 annual report later recorded 22,364 AI-related complaints and $893 million in adjusted losses, most of it investment fraud.

What does the FBI recommend?

A family secret word, calling back on known numbers, listening for inconsistencies, checking images for artefacts, limiting public exposure of your voice and face, and reporting to IC3.

Does the advisory name any victims or criminals?

No. It synthesises patterns across many complaints rather than describing a single case.

Sources

Internet Crime Complaint Center (IC3) | Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud (PSA241203). FBI Internet Crime Complaint Center (IC3) (primary). Primary source: full text of the December 3, 2024 advisory. Re-verified via direct fetch; fetched content matches all four AI-fraud modalities and the full tips list attributed to it in this record.
2025 IC3 Annual Report. FBI (primary). AI section and descriptor tables provide the 22,364 complaints and $893,346,472 adjusted-loss figures. Appendices B and C explain descriptor and complaint-data methodology. Reused the full cached report in this review; a prior missing worker excerpt did not make the PDF unavailable.
Cryptocurrency and AI Scams Bilk Americans of Billions. FBI (primary). Primary source: FBI press release, dated April 6, 2026, summarizing the 2025 IC3 Annual Report's AI-fraud figures. Re-verified via direct fetch.
Annual Reports - Internet Crime Complaint Center (IC3). IC3 (primary). Primary source: IC3's own hosting of the 2025 Annual Report referenced for total complaint/loss figures. Re-verified via direct fetch.

Cite this case

Social Engineering Examples. “FBI PSA241203: how criminals use generative AI in financial fraud (2024)”. Accessed 3 November 2026. https://socialengineeringexamples.com/fbi-ic3-psa241203-generative-ai-financial-fraud-2024

Attack Chain & Defense

1Impersonation material

What happened

the PSA describes criminals using synthetic profiles, images, identification documents, audio and video; it does not reconstruct one target-selection process.