ABSTRACT
It is generally understood that the principal mischief of deepfakes is the absence of consent; the depicted individual did not consent to this use of their digital likeness. In September 2025, OpenAI tested the limits of this account. It released a social media app called ‘Sora’ that allowed users to create and share a hyperrealistic likeness of themselves (a ‘deepfake’) using artificial intelligence. OpenAI claimed that all deepfakes generated using Sora were made with the consent of the depicted individual. In other words, the presence of consent conferred legitimacy on Sora-generated deepfakes, as distinct from the non-consensual deepfakes that have plagued the internet since 2017.
Drawing from consent theory and moral philosophy, this Article explains why OpenAI’s approach was flawed, and should not serve as a model for digital likeness generation. First, the consent provided by Sora users was neither informed, voluntary, nor contemporaneous. Second, even if these deficiencies could be cured, certain uses of our AI likeness may be fundamentally inconsentable, or incapable of legitimation through consent. Drawing on virtue ethics and feminist theory, this Article explains why hyperrealistic AI-generated content can inflict representational harm upon non-consenting third parties, even if the depicted individual has consented to this use of their likeness.
By unpacking the problematic features of consensual deepfakes, this Article shows that the legitimacy of AI-generated content does not turn on the presence or absence of individual consent. In fact, ‘consensual’ deepfakes share many of the same problematic features as nonconsensual deepfakes, including the impossibility of third-party consent, the deficiencies of platform-procured consent, and the vulnerability of ocularcentric institutions to the phenomenal immediacy of non-veridical representations. By exposing the limits of what consent can render legitimate, this Article enables app developers to more responsibly navigate the speculative terrain of digital likeness generation.
Geddes, Katrina, The Inconsentability of Deepfakes (July 14, 2026). Forthcoming, Harvard Journal of Law and Technology (2027).
Leave a Reply