MPs call for UK AI copyright law deadline as US legislation tabled

The recommendation was made by a committee of MPs in the UK parliament, which said content creators must be given control over whether their copyrighted works are used by AI developers to train their AI systems.

The report by the Culture, Media and Sport Committee comes after the UK government was forced to abandon efforts to broker industry agreement on a voluntary AI copyright code earlier this year, following a breakdown in talks. It was envisaged that the code would serve to strike a balance between AI developers’ desire to access quality data to train their AI models and content creators’ right to control – and commercialise – access to their copyrighted works.

The government is now in the process of engaging separately with representatives from the AI and creative sectors with a view to finding “a workable alternative” to a code – but the committee has said those talks should not go on indefinitely.

“Despite our previous recommendations that the government win back the trust of creators regarding their concerns over AI, its working group has not been able to bring forward a code of practice on AI and intellectual property,” the committee said. “Although the government asserted that it could consider legislating were agreement not reached, it has not indicated that it will do so. It is unlikely that simply conducting a further period of engagement with the sectors, with no clarity over its overall aims, will have any meaningful effect. We are concerned that the status quo simply favours AI developers, given creators’ concerns that their IP is already being used in AI development without licence or any practical means of recourse.”

“The government must ensure that creators have proper mechanisms to enforce their consent and receive fair compensation for use of their work by AI developers. It should set out measurable objectives for the period of engagement with the AI and rightsholders sectors, which it has said ministers will lead on, and provide a definitive deadline at which it will step in with legislation in order to break any deadlock,” the committee said.

There is the prospect of legislative intervention on the issue of AI development and copyright in the EU and in the US.

Under the EU AI Act, which is in the final stages of being formally approved by law makers, providers of so-called general purpose AI (GPAI) models would be obliged to put in place a policy to respect EU copyright law. The legislation states that such a policy will particularly need to account for reservations of rights pertaining to how copyrighted content that is mined is used. EU copyright law provides for such rights to be reserved by rights holders “in an appropriate manner” in the context of text and data mining by third parties, which EU copyright law provides for.

Providers of GPAI models would also be required to draw up and make publicly available a sufficiently detailed summary about the content used for training their model. The EU AI Office would be responsible for preparing a template to be used for making such disclosures. Recitals to the proposed AI Act make clear that while trade secrets and other confidential business information would not need to be disclosed in the summaries, providers would need to ensure there is sufficient information provided in them to enable copyright holders to exercise and enforce their rights. It is suggested in the legislation that the main data collections or sets that went into training the model, such as large private or public databases or data archives, be listed in the summaries and that providers share a narrative explanation about other data sources used.

In the US, one law maker this week introduced proposed new legislation before the House of Representatives which, if enacted, would impose similar disclosure requirements on developers of generative AI systems in respect of the data used to train their systems.

Under Adam Schiff’s proposed Generative AI Copyright Disclosure Act of 2024 (5-page / 153KB PDF), a person creating a training dataset used in building gen-AI systems would have 30 days to publish “a sufficiently detailed summary” of any copyrighted works used in the training dataset. An almost identical requirement would also arise for persons that alter a training dataset in a significant manner. If developers use a publicly available dataset, they would further be required to include a link to that dataset as part of their disclosure. The disclosures would have to be made in the US Register of Copyrights.

Schiff has proposed that fines of not less than $5,000 be imposed on persons that breach the disclosure requirements he has drafted.