CEO voice-cloning scams: the new threat draining French business accounts in 2026

The "fake president" scam is no longer a classic of B2B fraud: it has become a mass-market phenomenon supercharged by generative AI. In the first half of 2026, the French gendarmerie logged several dozen successful attempts at fraudulent wire transfers initiated by phone calls in which a manager's voice had been deepfake-cloned. For a few dozen euros and an online tool, a scammer can now reproduce the voice of a CEO, a CFO or a relative from a few seconds of audio posted on social media. For an SMS service like Texto SMS Gratuit, this is the right moment to remind everyone why human verification of an unusual call remains the only credible safeguard.

How a voice deepfake scam works

The method, documented by the "Cloned voices and generative AI" unit created in early 2026 by the gendarmerie, follows a four-step chain.

1. Voice harvesting: a few seconds of a manager are enough, captured during a public conference, a LinkedIn video, a podcast or a poorly protected voicemail. 2026 AI models no longer require long recordings.
2. Clone training: in less than an hour, a consumer-grade tool produces a synthetic voice capable of reproducing the victim's timbre, intonation and even speech tics. The result is sometimes indistinguishable from a real voice, even for a close relative.
3. Targeted call: the fraudster calls an employee authorised to make wire transfers (accountant, executive assistant, HR director) pretending to be the manager. They invoke an urgent, confidential operation to be completed before an unverifiable deadline.
4. Immediate transfer: the target, under the pressure of urgency and voice recognition, validates a transfer to a third-party account, often abroad. The amount can reach several hundred thousand euros.

What's new in 2026 is that the voice is no longer merely altered: it is cloned identically. Standard banking anti-fraud filters, which rely on geolocation or usual patterns, do not flag the danger, because the call does come from a phone number — often spoofing a French mobile number through CLI spoofing.

Why 2026 marks a turning point

Three factors converge to explain the current explosion. First, the democratisation of the technology: voice-cloning tools are available for less than €50, and some are free in a limited version. Second, the availability of the raw material: LinkedIn, YouTube, corporate podcasts and webinar replays are full of executive voices captured in excellent quality. Third, the industrialisation of fraud: cybercriminal groups now offer turnkey "CEO scam kits", with on-demand cloned voices, multilingual urgency scripts and ready-made mule bank accounts.

The numbers speak for themselves: the 33700 platform, the free hotline for reporting SMS and voice spam, has logged 6.2 million reports since its creation, of which around 5.1 million for SMS spam alone. Operator anti-spam filters — Orange, SFR, Bouygues Telecom and Free — block millions of messages every day before they reach terminals, but the voice channel remains far less protected than the text channel, as we pointed out in our article on the decline of SMS roaming in France.

Photo: Unsplash — a single call is enough today to trigger a fraudulent wire transfer of several hundred thousand euros.

The signals that should warn you

A call from your manager, an important client or a relative asking you to perform an urgent financial operation should trigger a series of systematic checks:

• Excessive urgency: any call imposing a deadline of a few hours, with a ban on talking about it, is a classic fraud marker. Real managers accept a callback.
• Unusual channel: if your CEO calls you on your personal mobile when they have never done so before, be extremely wary. This is also the pattern used in the DGFiP tax smishing scam: appearing through a channel that is not normally used.
• Request for secrecy: "don't tell anyone, it's confidential" is the most common opening line of scammers. A real urgent file can wait for verification.
• Unusual destination account: any transfer to an account whose country, bank or IBAN differs from those usually used with the caller must be blocked until verified.
• Slightly metallic voice: even the best 2026 deepfakes sometimes retain micro-hesitations, breath cut-outs or audible digital artefacts. If something sounds "off", hang up.

The verification procedure to apply

When faced with a suspicious call, apply the "Hang up, verify, call back" procedure:

1. Hang up immediately without validating any operation. Simply say "I'll call you back in five minutes" and cut the line.
2. Verify through another channel: call your manager on the number you normally use (and not the one displayed on screen, which may be spoofed). Send an SMS, an email or use your internal messaging.
3. Call back your official contact on the known number, to confirm they really made that call. If they are not reachable, wait until they are: no real emergency can resist 30 minutes of verification.
4. Document the incident: note the time, the displayed number, the reason given and any useful information. Report it immediately to your IT department, your bank and, in case of a successful attempt, to the Perceval platform of the French Ministry of the Interior.

The institutional network of the response

The fight against voice deepfakes now mobilises a full ecosystem. The French gendarmerie set up in early 2026 a dedicated "Cloned voices and generative AI" unit within its Centre for the Fight Against Digital Crime (C3N). Cybermalveillance.gouv.fr has published a guide dedicated to generative AI scams and offers an online diagnosis for victim companies. Arcep has opened in parallel an administrative investigation into spoofing, as we mentioned in our article on Arcep's investigations into number spoofing. Finally, the European Commission is working on a common framework to fight deepfakes, which could eventually impose mandatory watermarking of AI-generated content.

Operators, for their part, are testing real-time voice detection solutions: Orange announced in early 2026 a prototype capable of identifying a cloned voice in under 5 seconds with a 92% success rate on test corpora. Widespread rollout is not expected before 2027-2028, but the technology race has started.

Photo: Unsplash — detecting cloned voices relies on real-time analysis infrastructures still being rolled out.

What you need to remember

• Voice cloning now costs only a few dozen euros: trust in a voice is no longer a sufficient safeguard.
• Several dozens of successful attempts have been logged by the gendarmerie in the first half of 2026.
• Three reflexes: hang up, verify through another channel, call back on the usual number.
• Report any attempt to your bank, your IT department and the 33700 platform.
• The gendarmerie "Cloned voices and generative AI" unit now centralises the complaints linked to these scams.

To send a free SMS without giving your number to a third party and without leaving a trace, our sending form remains accessible to everyone, without registration. For any question on the privacy of your messages, you can consult our FAQ, our legal notice or contact us.