Legal contentions surround the fair use of copyright materials for training AI models. The question of whether AI-generated content can be copyright protected – often for marketing use – is under debate too. We look at the current status in some of Rouse’s main markets.
The country’s latest draft rules on Generative AI (under public consultation until July 2024) place the onus on AI providers to screen out copyrighted material from datasets used for training models.
Alongside this, the Guangzhou Internet Court ruled that Tab, a generative AI platform had infringed the copyright of ‘Ultraman’ works (February 2024). Part of the judgement set out the duty of care for AI providers to manage risks and establish reporting mechanisms for when copyright protected materials are identified in training data.
In terms of generated content, other Chinese court rulings have deemed materials created with sufficient human input can be copyright protected. This is potentially good news for brands and advertisers.
In the Springwind case (November 2023) the Beijing Internet Court ruled that an AI-generated image qualified for copyright based on the human author’s selection of prompt words, parameters and utilising AI as a creative tool. As such personal expression and originality was deemed to exist.
The first digital avatar case (April 2023) led the Hangzhou Internet Court to rule that the creators of an AI generated avatar could benefit from copyright. The avatar was generated from motion capture of a human, and it was argued software developer made unique aesthetic choices. The digital avatar itself was deemed exempt.
For companies using AI in China, we advise:
By contrast, United States’ regulators have yet to clarify their position on whether copyrighted materials can be used as training data. Companies have thus moved first.
The most prominent example is OpenAI signing licensing agreements with news organisations. The New York Times, however, has chosen not to enter into a deal, instead filing a lawsuit for infringement.
On copyrightability of content, an image generated by AI software and human prompts was recently deemed ineligible for copyright protection by the US Copyright Office.
The EU AI Act mandates the disclosure of training materials used in AI systems, with the aim of ensuring no copyrighted works are used without authorisation.
The extent to which AI-generated content can enjoy copyright protection remains undetermined. The EU Court of Justice has suggested that only human creators can be recognised as authors, but has not explicitly ruled out that AI-assisted creations can be considered original if they result from human creative choices.
In the absence of explicit legislation, companies using AI to generate marketing material should proceed with caution – especially if the content is critical to the brand. Be careful with using AI in generating logos, slogans, key brand imagery, etc.
United Arab Emirates. The Copyright Law, Federal Decree Law No. 38 of 2021 has been updated to include a range of protected works – such as smart applications and computer programs – though it does not explicitly address AI-generated works. It remains to be seen how evolving IP laws in the UAE will address AI’s impact on ownership, protection and enforcement.
Saudi Arabia. The Copyright Law, Royal Decree No. M/41 and its amendments do not explicitly cover AI-generated works. The Saudi Authority for Intellectual Property (SAIP) introduced a draft IP law (under public consultation) including AI related measures. It states that intellectual property created without human involvement will be public – implying no copyright protection.
Thailand. While the source code of AI may be protected as a literary work, the output generated by AI remains unprotected under Thai copyright law. AI is not recognised as a legal person capable of authorship.
Vietnam. The use of copyrighted works by AI systems – without the permission of the copyright owners – can constitute infringement if the use falls outside the exceptions of Vietnam’s Intellectual Property Law. Notably, these exceptions mostly cover personal or non-commercial purposes.
Indonesia. The Copyright Law is silent on protection of AI generated works. Therefore, Indonesia’s position with regards to this is uncertain.
Although the scraping in machine learning might be considered unauthorized reproduction under Indonesian Copyright Law, it will probably be difficult for a copyright owner to discharge the necessary burden of proof - because Indonesian judges would usually need to see tangible and original evidence of reproduction before it is willing to make findings of infringement. In the case of machine learning, a judge would be expected to arrive at some degree inference that reproduction must have taken place. Indonesian judges are probably not willing to reach legal conclusion based on inference.
Where AI specific laws, regulations and court precedents exist, it is critical to ensure your practices are compliant. In markets where specific legislation is lacking, Rouse recommends forming a clear understanding as to how traditional copyright law is applied, especially when creating marketing materials.
Regardless of location, best-practise principles can be implemented:
Transparency. The extent to which AI systems are used in creative works should be clear, so that AI’s impact on the finished content is evident.
Documentation. In case of a third-party challenge, make sure that any AI training and content creation processes are clearly documented.
Terms of Use. Review user agreements with AI services providers, paying attention to how the agreements relate to the ownership and commercial use of IP.
Authors: Holly White, Sunny Su, Yen Vu, Ly Nguyen, Theresa Mak, Gabrielle Chan, Kin Wah Chow, Uyen Doan, Matti Lindberg, Nadeen Helou, Ning (Nontaya) Chulajata