China bans use of deepfake technology without permission, or for “harmful purpose”

China’s Cyberspace Administration has published a new set of rules for the use of deepfake content, banning any productions made without the express permission of the protagonists, or that are used in a way that is contrary to national interests. The latter condition, ambiguously expressed, gives the authorities a free hand to arbitrarily censor any deepfake found on Chinese online, while imposing harsh penalties on the authors.

Relatively easy to create with your home PC or for a fee by accessing apps or services available online, deepfakes use artificial intelligence to create realistic representations – usually videos – of people saying and/or doing things they didn’t say and/or do in reality. A subject of controversy even outside China, deepfake technology is notorious for its potential to misinform and defame, with European authorities pressing tech giants such as Alphabet and Meta to step in to quickly remove abusive content. Companies that fail to comply could be fined up to 6% of global turnover.

By contrast, China’s internet is one of the most restricted in the world, “protected” by a firewall that blocks access to Western services and a censorship algorithm built into most social media platforms that automatically flags keywords and prevents messages from being sent. So the news of a de facto ban on deepfake productions is not surprising, as the measure only serves to limit the potential for dissent against the communist regime.

However, the way in which the new rules have been drawn up seems to recognise the importance and role of digital content synthesis technologies, including deepfake. In fact, a “telling” example was provided by a South Korean TV station that replaced its news anchor with an AI-animated deepfake version. Thus, China expects “synthetic humans” to be widely used in applications such as chatbots, news presentation, to movie productions that license the faces and physiognomy of well-known actors. In such scenarios, deepfakes should be marked as digital creations.

Also, deepfake creators – called “deep synthesis service providers” – must ensure that AI/ML models and algorithms render as accurately as possible the physiognomy and mimicry of the real person they simulate. At the same time, creators are responsible for the security of the data collected and can be held liable if it is misused. The rules also include a requirement for users to register – including their real names.