In early 2024, Hong Kong police reported that a finance worker at a multinational firm was deceived into authorizing $25 million in transfers after a video conference with deepfake recreations of the company's CFO and other colleagues. The attack combined social engineering with generative AI — a pattern since replicated in variations worldwide.
AI as attack vector, not just tool
Enterprises focus governance on their AI deployments. This case flips the lens: adversaries use AI against the enterprise. Synthetic voice and video erode trust in channels finance teams rely on for authorization.
Related reported incidents
- UK energy firm lost €220,000 to a deepfake voice call impersonating a CEO (2019, precursor trend)
- Multiple banks warned of AI-generated KYC document fraud in 2024–2025
- Enterprises report phishing emails indistinguishable from executive writing style via LLMs
Governance implications
Payment and treasury agents — including AI-assisted workflows — need stricter controls than customer support bots:
- Multi-factor authorization for transfers above thresholds, regardless of video or voice
- Out-of-band verification for instructions received through any digital channel
- Tool allowlists — payment APIs cannot be invoked by general-purpose agents without policy match
- Runtime deny on undeclared financial tool invocations with immediate GRC alert
Labeling and detection
China's deep synthesis rules and EU AI Act transparency duties push the opposite direction for defenders: know when media is AI-generated. Enterprises should combine technical detection with process controls — governance is not only about deploying AI safely, but surviving AI-enabled fraud.