Deepfake

Deepfakes use deep learning — typically generative adversarial networks (GANs) or diffusion models — to superimpose a person's likeness onto another person's body or voice in video. The original technique, which emerged around 2017-2018, used GANs to swap faces between video sources, requiring significant training data and compute but producing increasingly convincing results. Modern deepfake techniques have become substantially more accessible: consumer applications can swap faces in video with minimal technical knowledge, voice cloning can reproduce a person's voice from seconds of sample audio, and full-body synthesis can generate realistic video of a person doing or saying anything from a short source video. The quality frontier continues advancing with diffusion-based approaches that exceed earlier GAN results.

The ethical and legal landscape of deepfakes is complex and rapidly evolving. The same technical capability underlies legitimate creative uses (entertainment, satire with disclosed AI modification, consensual avatar creation) and harmful ones (non-consensual intimate imagery, political disinformation, fraud, harassment). The distinguishing factors are consent (does the depicted person know and agree to the use of their likeness?), disclosure (is the AI-generated nature of the content clearly communicated to viewers?), and intent (is the content designed to deceive, harm, or manipulate?). Multiple jurisdictions have enacted or are considering legislation specifically targeting malicious deepfake creation, particularly non-consensual intimate deepfakes and politically deceptive deepfakes in election contexts.

For B2B teams, the relevant frame for deepfakes is not the harmful use cases but the governance questions that arise from the same underlying technology being used in legitimate commercial contexts. AI avatar creation, voice cloning for localization, and synthetic presenter creation all involve the same deep learning techniques as problematic deepfakes — the difference is consent, disclosure, and purpose. This means B2B teams using these technologies should: obtain explicit written consent from any person whose likeness or voice is used in AI-generated content, maintain clear internal policies on approved use cases, implement disclosure practices for AI-generated content in customer-facing applications, and recognize that audience trust depends on how transparently the technology is deployed.