Deepfakes and Intellectual Property: Copyright licensing, personality rights, and civil/criminal liability for synthetic media generation

Abstract

Background. Generative AI now makes realistic synthetic video, images and voices of real people cheap and widespread. Law must deter non-consensual deepfakes while allowing a lawful market in licensed digital replicas, but copyright, personality rights and criminal law each address only part of the problem.

Objective. To compare how legal systems allocate control over a person’s likeness and voice, govern licensing, and distribute civil and criminal liability for synthetic media.

Methods. A doctrinal, functional comparative analysis of eight legal systems (the EU layer, Italy, the United Kingdom, the United States, China, South Korea, India and Bangladesh) as of 20 October 2026. Each system was scored on a nine-dimension Synthetic Media Legal Coverage Index (SMLCI, 0–18) grouped into three pillars: rights and licensing, transparency and platforms, and liability. Four IP-style likeness-right models (Denmark, Cyprus, the U.S. NO FAKES bill and Tennessee’s ELVIS Act) were compared separately.

Results. SMLCI scores ranged from 6 (Bangladesh; 33.3%) to 15 (United States; 83.3%), with a mean of 9.9. The rights and licensing pillar reached only 32.8% of its maximum, against 75.0% for transparency and platforms and 70.8% for liability. Sexual-deepfake offences were the most widely covered dimension (87.5%), while copyright or IP vehicles for likeness (6.3%) and post-mortem control (25.0%) were the least. Rankings were robust to equal pillar weighting (Spearman ρ = 0.99).

Conclusion. Deepfake law has built sanctions and takedown duties faster than the property and licensing rules that define lawful use. Copyright is best used as an enforcement container for a dignity-based personality right, not as its foundation. A four-layer model (right, licence, transparency, liability) is proposed for jurisdictions such as Bangladesh.