The rapid development of generative artificial intelligence has intensified concerns about AI-driven manipulation. This chapter examines one of its most controversial applications, namely deepfake-technology, which enables the creation of highly realistic audiovisual content. While existing theories of manipulation correctly identify deceptive deepfakes as manipulative, they often overlook a crucial dimension: the structural epistemic risks embedded in the technology itself. This chapter shows that distinguishing between benign and malicious deepfakes provides only a partial account of deepfake manipulation if this distinction is not connected to the epistemic harm deepfakes may cause. The central claim is that malicious deepfakes are manipulative not only insofar as they are created and disseminated to deceive, but also because they transform the epistemic risk inherent in deepfake-technology into concrete epistemic harm.