Can an AI clone a voice or a face well enough to scam you?

Yes: AI can now generate voices and faces convincing enough that scammers have used them to steal real money. Deepfakes are images, videos or audio edited or generated using artificial intelligence, and audio deepfake technology, also called voice cloning, is designed to generate speech that convincingly mimics specific individuals.

The first widely reported cases targeted companies. After a much-cited 2019 case in which a cloned chief executive's voice won a €220,000 transfer, the same technique impersonated a company director in early 2020 in a scheme that convinced a branch manager to transfer $35 million. Later the firm Arup fell victim to a deepfake scam at its Hong Kong office, losing approximately $25 million, when fraudsters used AI-generated video and audio to impersonate senior officials and deceive an employee.

For ordinary people, the worry is a call from a "relative". As of 2023, voice cloning from a few seconds to a minute of recording, combined with new text generation tools, enabled automated impersonation scams using a convincing digital clone of a friend or relative. In March 2023 the US Federal Trade Commission warned consumers about AI faking the voice of a family member in distress asking for money. According to a 2023 McAfee survey, one person in ten reported being targeted by an AI voice cloning scam, and 77% of those targets reported losing money.

The psychology is the old one. Scammers create urgency or a fear of authority, and the AI simply makes the impersonation more believable.