Intesa's Fideuram lost €95 million to a fake CEO and a reported deepfake

Fraudsters moved €95 million out of Fideuram, Intesa Sanpaolo's private banking arm, in a deepfake CEO fraud, Reuters and Corriere della Sera's Milan edition reported on 25 September. In February, per Reuters' sources, the then-chairman got a WhatsApp message, seemingly from group CEO Carlo Messina, asking for urgent help with an international payment. A follow-up call, apparently from a senior partner at a major law firm, confirmed it. The callers reportedly used AI to reproduce the lawyer's voice. The scheme worked on trust, not systems.

Payments went to accounts mainly in China and Hong Kong. Banks and authorities in China, Portugal and Italy recovered about €53 million; roughly €36 million is still missing, exchanged for cryptocurrency, according to Reuters' sources. One of those sources says Milan prosecutors are investigating a foreign national living outside Europe on suspicion of computer fraud; the sources say no Fideuram executive is under investigation. Corriere della Sera reports emails posing as the law firm; La Verità counts 11 documents, including a power of attorney purportedly signed by Messina.

Each layer vouched for the one before, all aimed at the person with authority to move money. In our view, an authorization check can confirm that the right person approved a transfer; it cannot tell whether that person was deceived. Deepfake audio is not rare: in a Gartner survey of 297 senior cybersecurity leaders, 41% reported at least one social-engineering incident involving a deepfake during an employee audio call in the past 12 months, per Help Net Security.

• Make out-of-band verification the default for payment instructions arriving by messaging app or voice, however senior the requester.
• Map whose voices, relationships and signatures are publicly exposed, and brief those people first.
• Pre-agree the recovery playbook: who contacts which banks and authorities abroad, and who signs off.

OBSIDIAN — Counterintelligence is designed to give early warning when someone is targeting an organization's people, not just its systems. It maps what the organization protects to who wants it and watches for targeting indicators across open-source, cyber and travel signals. Program protection and security staff validate each warning and choose the countermeasures; at threshold, a concern opens an auditable case with the evidence attached.

The industry warning on AI-enabled attacks described a capability. Fideuram shows its price once it reaches someone with payment authority. The defense is procedural before it is technical: verification on a channel the attacker cannot supply, a map of the people impersonators will copy, and an investigation ready to move before the money does.