As generative AI continues to transform the entertainment industry, there is an increasing demand from AI developers for access to rights-protected content that can be used to “train” AI foundational models. Content owners are capitalising on this demand by entering into licensing deals with AI companies – granting them rights to use the content for AI training purposes and thereby exploiting the content in new ways that have not been contemplated until now.
An urgent conversation is therefore needed to ensure that any such new exploitation of rights-protected content in the context of AI is carried out with full respect and recognition of performers’ property rights and applicable data protection laws.
This conversation must include production companies, broadcasters, streaming platforms, studios, videogame and audio publishers, videogame developers and vendors, record labels, advertising agencies, brands and any other entities that own or control content within which performers’ rights subsist and where performers are identifiable.
1. Background
Most foundational AI models used and commercialised today have been trained on vast quantities of high-quality (often rights-protected) materials, which have been crawled and scraped from online sources through a process known as “text-and-data mining” (TDM).
Much of this TDM and AI training has been carried out without consent from - or remuneration for - creators (as copyright owners) and/or performers (as performers’ rightsholders and data subjects). AI development companies and data aggregators have therefore profited from enormous reserves of works created by generations of union labour, without any buy-in or recognition towards the people that created those works.
Whilst the UK government is currently consulting on whether to update domestic intellectual property laws to allow for the commercial TDM of rights-protected works by AI companies (unless the content owner has effectively “opted out”), the position under current law is clear. Currently, TDM is only permitted as an exception to copyright and/or performers’ rights in the UK under very narrow conditions – there must be "lawful access" to the work and the purpose of TDM must be restricted to non-commercial research only. This means, to the extent TDM activities have already been carried out by commercially-motivated AI companies without rightsholder consent, this constitutes a breach of intellectual property rights.
As a result of this situation, we understand that some AI companies are now approaching content owners within our industry to negotiate licensing deals which grant access (or authorise prior access) to media content for the purpose of training AI models. In effect, these technology companies are attempting to retrospectively authorise their past infringements of works created by our members by securing licences to the copyright in those works from content owners, but without due consideration to performers’ rights subsisting within those same works.
Moreover, we are aware that some content owners are starting to develop their own foundational AI models (either independently or in partnership with other organisations) using training datasets that include content featuring performers’ contributions.
In light of this, it is important that we are clear on the existing and longstanding rights that performers have in relation to content in which they feature and are identifiable.
2. Performers’ rights under the Copyright, Designs and Patents Act 1988 (CDPA)
Performers’ rights are recognised under Part II of the CDPA both as forms of “property” and “non-property” rights. A key performers’ property right includes the “right of reproduction”, being the right for performers to control who is able to record and make reproductions of their performances.
The UK government has made clear that the right of reproduction under the CDPA covers all technology, including AI. It is widely acknowledged that both the processes of (1) compiling machine learning AI datasets through TDM; and (2) subsequent AI training using those datasets, involve the making of 'copies' and/or ‘reproductions’ of works within the meaning of the CDPA. These processes therefore require the express consent of relevant copyright owners and performers (unless such rights have been effectively obtained under contract or a copyright exception applies).
Performers’ rights can be transferred to third parties (either via assignments or licences) or otherwise ‘waived’ by performers under mutually agreed contracts. Until now, most contracts have required performers to give transfers or waivers of their performance rights either:
• in a very broad and generic context, without specific detail in the underlying contract on the intended forms of exploitation of those rights and resulting works; or
• in the context of specific commercial purposes contemplated or disclosed to the performer at the time, such as for the making of a film, TV programme, videogame, commercial or radio show (with payment structures to reflect such usage rights).
In either case, such contracts were not drafted to envisage the exploitation of performers’ rights for AI training purposes and no provision has been made for this use case in the underlying contract, since this form of exploitation was simply not in contemplation of the parties at the time of drafting, negotiating or agreeing the contractual terms.
Therefore, it is Equity’s position that any contract (including those currently construed under Equity collective agreements) in which creative workers have consented to transfers or waivers of performance rights (either broadly or in the context of specific commercial purposes) should not be interpreted as a legal basis for exploiting such performance rights in the context of new technologies that were not contemplated at the time of the contract, such as for commercial TDM, AI training or digital imitation purposes, unless the contract explicitly references those activities as an authorised form of exploitation.
In the absence of a relevant commercial TDM exception under the CDPA or an express contractual provision allowing for the exploitation of performers’ rights for AI training purposes, any act of TDM and/or AI training using rights-protected content for commercial purposes constitutes an infringement of our members’ performers’ rights.
Furthermore, irrespective of whether a new commercial TDM exception is introduced in the UK following the government’s ongoing consultation, any such exception would only serve to authorise commercial TDM activities on material that has not been “opted out” by the relevant rights holders when deployed to collect ‘input’ material for the training of AI models. However, any such commercial TDM exception would have no impact on any resulting ‘output’ of AI models, which may separately infringe intellectual property rights (including performers’ rights) to the extent such output is deemed to have copied or reproduced rights-protected works without due authorisation.
3. Data protection rights under UK General Data Protection Regulation (GDPR)
Uses of content for AI purposes also engage performers’ rights as data subjects under the UK GDPR. These data protection rights are engaged wherever a performer’s voice, face or other identifying features are captured, used and “processed” by third parties (and “processing” includes the “sharing” of personal data).
The processing of performers’ personal data by third parties – including content owners (who may wish to share performers’ personal data with AI companies under a licensing deal) and/or AI companies (who may wish to collect and store performers’ personal data when training AI models) – requires those third parties to establish a lawful basis for such processing under the UK GDPR.
One such lawful basis is where the ‘consent’ of the data subject has been obtained. For the same reasons as set out above, it is Equity’s position that performers who have been asked to give transfers or waivers their performance rights under contract (either broadly or in the context of specific commercial purposes) have clearly not given their free, informed and specific consent to the processing of their data for AI training purposes (and, in any case, such consent can be withdrawn by performers).
Another such lawful basis is where the processing is “necessary for the performance of a contract” that the processing entity has in place with the data subject. Since AI companies have conducted mass-scale TDM without any agreements in place, or generally only contract with content owners under licensing deals, there is no contract in place with data subjects for AI companies to rely on for processing data for AI training purposes.
As for whether a performer’s contract can be relied upon by a content owner as a lawful basis for processing (e.g. by sharing personal data with an AI company under a content licensing deal), the key questions is whether the processing is truly necessary for the performance of the contract. The UK’s Information Commissioner’s Office (“ICO”) has made clear in its recent response to the consultation series on generative AI that: "For the creative industries, the contract lawful basis is very unlikely to apply as it is unlikely that an organisation is under a contractual obligation to use a creator’s content to train its generative AI."
Equity firmly stands behind this position, especially where broad and generic performer consents to processing have been bundled-in with other contractual terms. Accordingly, where a performer’s contract is for making a film, TV programme, videogame, commercial, or a radio show, such a contract should not constitute a lawful basis under data protection law to use this performance data to train an AI model or generate a digital replica. This is true even if the contract wording includes a general ‘release’ of a performer’s data rights, because AI uses are not the real purpose of the original contract.
Finally, the UK ICO has also made clear in its consultation response that the often-relied-upon “legitimate interests” lawful basis will be interpreted narrowly in the context of web scraping to train AI models, and is unlikely to apply to widescale TDM by AI companies where alternative data collection methods may be feasible.
As such, Equity asserts that the processing of its members’ personal data for AI purposes has, up to this point, been carried out without a lawful basis and in breach of the UK GDPR. Any breach of data protection rights can give rise to legal claims and be subject to regulatory sanctions and (in serious cases) financial penalties. Equity will be ensuring our members’ rights in this area are respected and hold data controllers to account for any unlawful processing of our members’ data. A more detailed explanation of our position on GDPR can be found on our website.
4. Next steps
We invite engagers across the audio and audio-visual industries to work constructively with us to establish fair and legally compliant licensing and data protection frameworks for AI. We have had productive conversations with companies who engage our members for the express purpose of creating content to train foundational AI models, with the appropriate remuneration and permissions expressly granted for these uses. Equity hopes to have similar transparent and open conversations with others who engage our members.
Whilst we support a growing ecosystem of data use, in view of the rights outlined above, consent must be obtained directly from our members or via the union on a collective basis before any use of relevant content for AI purposes. Moreover, our members must be compensated in any future licensing arrangement with AI companies. These remuneration schemes will need to cover (1) past data mining, technology development and use of their work, which took place without appropriate consent or permissions, and (2) future use of their work in this way. Crucially, these remuneration schemes should be voluntary and our members should be free to join or abstain from participating.
Where performers’ rights are breached, Equity will robustly defend members, including via the courts if necessary. We look forward to a constructive dialogue with producers to ensure historical, current and future use of our members’ work is based always on informed consent, transparency and compensation.
Yours sincerely,
Paul W. Fleming
General Secretary, Equity
Please read our guidance page for a more detailed explanation of our position on GDPR