AI‑Powered CEO Voice Spoofing Scams Steal Over $100 Million, DOJ Indicts Three
What Happened — The U.S. Department of Justice unsealed an indictment against three men—Jacob Soall, Sharon Hazan and Guy Sheetrit—for running an AI‑enabled “CEO‑spoofing” fraud campaign that extracted more than $129 million from companies worldwide between 2024 and 2026. The conspirators used generative‑AI tools to synthesize the voices and images of senior executives and external lawyers, contacting lower‑level employees and convincing them to wire large sums to accounts under the scammers’ control. Victims included a Missouri hospital ($4.9 M), a Rhode Island bank ($2.6 M), a California hospital ($2.9 M), a U.K. tech firm ($11.5 M), a Massachusetts finance company ($4.6 M) and a Florida pet‑food company ($5.2 M). Funds were moved through U.S., European, UAE and Chinese banks before being converted to cryptocurrency. All three suspects have been arrested abroad; extradition requests are pending. Source: The Record
Why It Matters for Trust & Control Assurance — The episode shows that AI‑generated deepfakes can bypass traditional verification controls, making robust security‑awareness programs and documented payment‑approval processes essential.
Who Is Affected — Healthcare providers, banks, technology firms, finance companies, and other enterprises that process large wire transfers.
Recommended Actions
- Institute a multi‑factor, out‑of‑band verification step for any executive‑initiated payment request.
- Run regular, AI‑deepfake‑focused security‑awareness simulations for staff handling financial transactions.
- Update policies to require independent confirmation (e.g., via a known corporate channel) before acting on urgent payment instructions. Source: The Record
Technical Notes
- Attack vector: AI‑generated voice and image spoofing (social engineering).
- No software vulnerability disclosed; the threat leveraged generative‑AI to create convincing impersonations.
- Financial loss: > $129 million across multiple sectors. Source: The Record