A deepfake is AI-generated or manipulated image, audio or video content that looks like real persons, places or events and could be mistaken for genuine (Art. 3(60) AI Act). Particularly relevant for organizations are cloned voices and faked video calls in which attackers, for example in the name of senior management, demand an urgent payment.
Deployers who generate deepfakes with an AI system disclose under Art. 50(4) that the content has been artificially generated or manipulated. For evidently artistic or satirical works, a lighter duty applies. Providers of generative systems additionally mark outputs in a machine-readable format (Art. 50(2)). Call-backs to known numbers, four-eyes approval and training help against fraud. On missing disclosure, see MU-13 Presenting outputs as human work, and on the fraud route, G 0.42 Social Engineering.