- → The Banque de France alerted in April 2026 about video deepfakes imitating its governor François Villeroy de Galhau
- → The DGSI has identified a case of videoconference scam where the face and voice of a manager were cloned
- → The volume of online deepfakes increased tenfold between 2023 and 2025, now several million pieces of content
- → 49% of global companies say they have already experienced an attempted fraud by deepfakeSynthetic content (video, audio, image) created by AI to realistically imitate a person, with risks of fraud and misinformation. audio or video
Cloning a voice requires 30 seconds of audio recording. Creating a credible video of a CEO via video conference takes an afternoon for a well-equipped con artist. In 2026, AI scams have moved from the laboratory to mass production, and they affect savers as much as large companies. The Bank of France sounded the alarm in April, the DGSI published three real cases of French companies trapped. Overview of what everyone needs to understand to avoid falling into the trap.
01 · The peak
The volume of deepfakes has increased tenfold in two years
The figure comes from a study by the company Regula taken up by the DGSI in its January 2026 interference flash: the volume of deepfakes circulating online was multiplied by more than 10 between 2023 and 2025. Today there are several million of them. Most of it is entertainment (parodies, films, creative content), but a growing part, and this is the problem, is directly aimed at defrauding.
On the business side, the barometer is edifying: 49% of global organizations acknowledge having already been targeted by an attempted deepfake audio or video scam. One out of two. And only a fraction of attempts are publicly reported, because victims prefer to avoid scandal.
×10 · deepfakes volume 2023-2025
49% · targeted companies
30 seconds · audio is enough to clone a voice
2026 · France moved to official alert level
02 · Bank of France
The official alert: a fake Villeroy de Galhau recommends an investment to you
On April 27, 2026, the Banque de France and the ACPR (Prudential Supervisory Authority) launched a joint alert on videos circulating online. We see Governor François Villeroy de Galhau, his face, his voice, his intonation, recommending a “highly lucrative” investment to savers. Except it’s not him. It’s an AI-generated deepfake.
The same operation also targeted Emmanuelle Assouan, secretary general of the ACPR. Scammers rely on notoriety and institutional credibility: if “the Bank of France” recommends something to you, you naturally lower your guard.
Banque de France officials never recommend any investment. It feels better saying it.
Official press release from the Banque de France and the ACPR (April 27, 2026)
The typical scenario is well-established. You see the video on Facebook, Instagram or TikTok. You click. You are asked to register to “access placement”. A few days later, a fake advisor calls you, tells you about an upcoming large transfer, and asks you to pay “unblocking fees” or to carry out “banking transactions on telephone instructions”. Money lost, phantom victim, scammer already gone.
03 · DGSI
The DGSI case: a CEO cloned by videoconference requests a transfer
The domestic intelligence service published a flash of economic interference in January 2026 which is worth reading in full. Among the three cases exposed, one became a textbook case for security departments.
An industrial site manager of a French group receives a call by videoconference. On the line, the leader of the group. Perfect face, exact voice, credible intonation. The call has one objective: to make an urgent transfer as part of a “discreet acquisition” of the group. The manager, surprised by the unusual nature, ends the exchange and alerts management through the usual channels. Confirmation: it was a deepfake.
This case illustrates the new frontier of presidential scams. Before AI, you needed an actor who roughly imitated the voice over the phone, easily detectable. Today, we have a face animated in real time, a voice cloned with 30 seconds of public audio, and a credible background (LinkedIn to reconstruct an office, archive photos for details). All for less than 200 euros of tools accessible to anyone.
04 · How they do it
Mechanics are now industrial. Scammers no longer need to be technical experts: everything is self-service on the web, sometimes free, sometimes for a few dollars.
Voice cloning 30 seconds of audio (a podcast, a YouTube video, a stolen voicemail) is enough to train a voice model that can then say whatever you want.
Real-time face Tools open sourceSoftware whose source code, and sometimes the model weights, are freely published and reusable. allow you to put a target face on the scammer’s webcam. There latencyThe time elapsed between sending a request and receiving the full response from a model. has passed under 200 ms, the illusion holds up in videoconferencing.
Custom spear phishing The AI sifts through LinkedIn, company sites, and the local press. She prepares an ultra-targeted message that mentions a real project, a mutual colleague, a recent event.
The fake institutional site To trap savers, we clone an official site (Banque de France, ACPR, consumer bank) with a slightly modified domain name. The registration form is the gateway to the scam.
05 · The signs
The 5 signs that should alert you
Even in 2026, deepfakes are not perfect. Here’s what to watch out for, whether you’re a saver facing a Facebook video or an executive facing an unusual call.
1
The abnormal emergency
Any request that imposes a very short deadline (“you have to sign this evening”, “the transfer must go out before 5 p.m.”) is a red flag. Real institutions and real leaders don’t work like that.
2
The detail that is wrong in the video
Look at the eyes (irregular blinking), the lips (sometimes laggy lip syncing), the edges of the face (artifacts at the hairline). At low resolution, it’s harder, but it’s still detectable.
3
The unusual channel
A CEO who requests a transfer via WhatsApp, a minister who recommends an investment on TikTok, your banker who calls you from a foreign number: none of these channels are legitimate. Call back using the official number.
4
The “guaranteed return” content
No serious financial institution promises guaranteed returns, even less through video on social networks. If you are offered 15% without risk, the scam is statistically certain.
5
The request for confidentiality
“Above all, don’t tell anyone about it, it’s confidential. » This phrase is the most reliable sign of a scam. Real operations never require hiding information from your spouse or your accounting department.
06 · Reflexes
What to do if you encounter a suspicious case
Three simple reflexes allow you to neutralize almost all current deepfake scams, even the most sophisticated.
The first is the callback via official channel. Whatever the message received (video, call, email), you hang up and call the person or institution back using a number you already know, not the one you have just been given. If it’s your CEO, you call him on his regular cell phone. If it’s your bank, you call the number on the back of your card.
The second is the cross check. For any unusual financial request, ask for confirmation via a second channel (internal professional email, face-to-face, SMS on the known cell phone). For placements presented on video, look for the official press release on the institution’s website, it will be published if the offer really exists.
The third is the reporting. The Banque de France and the ACPR encourage people to report fake videos directly on Pharos (the official platform) and send them to them via their contact form. The DGSI recalls that it has an open communication channel for any suspicious events linked to AI. Reporting protects others and allows institutions to remove content faster.
07 · Tomorrow
2026-2027: what happens next
The rise will continue, and probably accelerate. Three trends will structure the subject in the next 12 months.
First, voice call scams will become commonplace. Cloned voices no longer require complex hardware: a simple desktop computer and a public audio sample are enough. Scammers will target individuals by posing as a loved one in distress (“Mom, I had an accident, send 500 euros right away”).
Then, European regulators will tighten their framework. L’AI ActThe 2024 European regulation which regulates the development and use of AI according to risk levels. already provides for reporting obligations for deepfake content, but the application is in its infancy. The platforms (Meta, TikTok, X) are starting to automatically label certain content, without yet covering everything.
Finally, technical detection will progress. Several French startups (including botSEO and specialized players monitored by NewsIA) are working on deepfake detection tools adapted to businesses. The fight will become a classic cat-mouse: generators more and more credible, detectors more and more refined. With, in the middle, the general public and the services that need to be trained.
To go further
To understand the tools behind voice cloning, our guide to the AI audio ecosystem covers players ElevenLabs, Suno and their competitors. For managers and accounting services exposed to presidential scams, our regulation and ethics category follows developments in the French and European framework. And to help your company train its teams, the free AI comparison lists the serious tools you need to know. On the same mechanisms, our investigation into AI disinformation targeting France and our article on cheating in the baccalaureate boosted by ChatGPT complete the picture.

