Twitch is a live-streaming service, owned by Amazon, where individuals broadcast content and can communicate with their viewers. Recently, Twitch faced criticism from users for feeding streamers’ content to train Amazon’s AI models.
Issues
Earlier this month Amazon disclosed a new feature allowing users to opt out of the default setting that permits their content to be used for AI training. This sparked concern and criticism from Twitch users.
Firstly, the disclosure prompts the question of how streamer content was previously used. As Twitch’s chief product officer, Mike Minton, acknowledged, it is unlikely that users would choose to opt in to have their content used as AI training data. Furthermore, the existence of a default setting may be problematic because it assumes some level of possession over streamer content without the explicit consent of individuals. Arguably this is an assumption against the autonomy and control of streamers over their content.
There is also the practical concern regarding the scenario in which a streamer’s content is rebroadcast on another account or platform that has not applied the opt-out setting. This could be addressed by making a notification claim through Twitch for the content to be removed from other accounts, although the content may already have been fed to an AI model by that point.
Given that AI models are designed to learn from and mimic the information they are given, the default setting is further problematic because it commodifies the voices and styles of creators, potentially raising intellectual property concerns. More theoretical questions may also arise regarding the difference between feeding an AI model content to learn from and allowing a person to view content that informs their own output.
Alternatively, there is the argument that streaming makes the content public and that a reduction in an individual’s control over the content they generate is therefore an inevitable reality. Mary Kish, head of community at Twitch, has stated that content collection for the development of AI has become the industry standard. Indeed, Meta uses public content (meaning content from Facebook and Instagram accounts that are set to public) to train its AI models. On this view, Twitch’s disclosure could be seen as a positive step towards transparency, giving streamers more control over their content through knowledge and the ability to opt out. However, the question remains as to whether Amazon was previously using Twitch streamers’ content to train its AI models.
Law
The question of how Amazon previously used Twitch streamer content has been highlighted by Twitch streamer Warren Pandiscia, who has suggested that Amazon had been using streamer content as early as 2024. Pandiscia is suing Amazon in the US for breach of contract, unjust enrichment and unfair business practices, on the basis that Amazon has allegedly used content without streamer consent. The lawsuit is a class action seeking damages on behalf of millions of Twitch streamers who allege that their content has been used by Amazon to train AI models without their consent. These allegations have not yet been determined by a court.
Beyond the contractual considerations, intellectual property and data protection law may also be relevant.
Copyright
Firstly, Twitch streamers might be able to claim under copyright law. In the UK, copyright protection arises automatically for original literary, dramatic, musical and artistic work (including illustration and photography), original non-literary written work (such as software, web content and databases), sound and music recordings, film and television recordings, broadcasts, and the typographical arrangement of published editions of written, dramatic and musical works.
Therefore, a Twitch streamer’s content could be automatically protected by copyright if the expression of the idea is deemed ‘original’. However, the originality requirement makes it harder, and thus less likely, for all streamed content to attract protection. It is also notable that under US law, while copyright protection is automatic, a person must register with the US Copyright Office before bringing a claim for infringement. In the UK, copyright is not registered.
Data Protection
The second possible option would be a claim under data protection law. However, the difficulty is that data protection legislation only protects personal data which, under s.3(2) of the Data Protection Act 2018, “means any information relating to an identified or identifiable living individual (subject to subsection (14)(c)).” Much of a streamer’s content is unlikely to fall within this category; while public posts and broadcasts may still contain personal data, a large proportion of the content would likely not qualify.
How can we help?
Taylor Hampton has extensive experience in intellectual property disputes, data protection and internet and social media law. If you are a content creator concerned about how your work is being used, please contact us on 0207 427 5970 or at enquiries@taylorhampton.co.uk.
Disclaimer: This article provides general guidance only and does not constitute legal advice. Legislation, civil procedure rules, and case law can change. Always seek professional legal advice tailored to your specific situation before acting.