AI Cloned One Voice and a Bank Sent €95 Million. Here's How It Happened

In February 2026, the chairman of Fideuram, Intesa Sanpaolo's private bank, got an urgent WhatsApp message that appeared to come from the group's CEO, then a phone call from a well-known lawyer confirming a secret deal. The lawyer's voice was an AI clone. Around €95 million left the bank in three days. No system was hacked. AI Hack Files #2: how the scam unfolded, how much was recovered, why voice is no longer proof, and 3 things to do so a cloned voice never gets you. ⏱ Chapters (adjust after edit) AI voice cloning just cost a bank 95 million euros. Here is how the attackers bypassed security with a simple phone call. This wasn't a complex system breach or a software exploit. Instead, attackers used sophisticated social engineering to trick employees into authorizing a massive transfer. This video breaks down how the fraud worked, why AI security protocols are failing to account for human manipulation, and the specific signs to watch for when you receive an unexpected call. We look at the anatomy of this bank fraud and how to protect yourself or your organization. You will learn the exact methods used to execute this voice scam and the simple defense strategies—like family safe words or multi-person verification—that can stop these attacks in their tracks. Subscribe for weekly security breakdowns, and let me know in the comments: what is your team's current policy on verifying wire transfers? ━━━━━━━━━━━━━━━━━━ 🔐 About Overprivileged The security side of AI, explained properly: what AI tools can actually access, what can go wrong, and what to do about it. ⚠️ Based on public reporting. The investigation is ongoing. Visuals are illustrations. Not financial or security advice. 0:00 The €95 million phone call 0:25 The WhatsApp from the "CEO" 0:55 The cloned voice 1:30 Three days, €95 million 1:55 No hacking required 2:20 What was recovered 3:15 Why this matters to you 3:50 What to do 4:35 Would you catch it? 📚 Sources (as of 9 Oct 2026) Tom's Hardware Italia: tomshw.it/altro/95-milioni-rubati-con-una-voce-clonata-dallia-come-e-stata-ingannata-una-banca-italiana The Cryptonomist: en.cryptonomist.ch/2026/10/02/ai-voice-scam-italy-fideuram/ Technology.org: technology.org/2026/09/28/fideuram-intesa-sanpaolo-ai-voice-scam-95-million/ TechNewsHub: technewshub.co.uk/post/deepfake-audio-and-spoofed-messages-trick-italian-private-bank-into-wiring-95-million-overseas Escudo Digital: escudodigital.com/en/cybersecurity/voice-deepfake-heist-95-million-stolen-from-italian-bank.html Original reporting: Corriere della Sera, Milano Finanza ⚠️ Based on public reporting as of 9 Oct 2026; figures vary slightly between reports. The investigation is ongoing and no one has been charged. This video explains what happened and how to protect yourself, not how to do it. Visuals are illustrations; no real voices are used. ━━━━━━━━━━━━━━━━━━ 🔐 About Overprivileged The security side of AI, explained properly: what AI tools can actually access, what can go wrong, and what to do about it. 📂 Series: AI Hack Files, real AI-powered attacks explained from public reports. ▶ AI Hack Files #1: One Hacker Used AI Agents to Breach South Korea's Banks 📱 Shorts on Instagram @over.privileged · TikTok @overprivileged.ai 💼 Business enquiries: overprivileged.lab@gmail.com #AIScam #Deepfake #VoiceCloning #Cybersecurity #BankFraud #AIHackFiles