EU considers copyright reform for gen-AI as US court rejects ‘fair use’ claim

Cerys Wyn Davies of Pinsent Masons, who specialises in AI and copyright, was commenting after the European Commission indicated that it is considering copyright reform to address challenges linked to the use of copyright-protected content for generative AI.

The Commission has invited content creators, AI developers, and other stakeholders to indicate the steps they think should be taken to address the challenges, with “new binding measures under EU copyright law”, enhancement of voluntary cooperation via “soft law instruments”, and non-legislative measures to support access and use of copyright-protected content for AI, among the options listed for intervention.

The Commission’s consultation, which runs until 3 November 2026, was opened on the day the US Court of Appeals rejected ‘fair use’ arguments in a dispute between media business Thomson Reuters and Ross Intelligence Inc.

Ross Intelligence built a legal research search engine that used AI. The product used training data that contained headnotes published by Westlaw. Those headnotes summarise key points of law and details of court determinations on them. Thomson Reuters, which owns the copyright for Westlaw works, previously refused Ross Intelligence a licence to use its content to train its rival product and claimed that Ross Intelligence was responsible for copyright infringement by using the Westlaw notes in the training of its product without permission to do so.

In a judgment last week, the US Court of Appeals ruled (32-page / 363KB PDF) that the Thomson Reuters material was copyrightable and that the use of it by Ross Intelligence was not covered by the ‘fair use’ limitation in US copyright law. The case is just one of many that have been litigated globally on questions of copyright that concern AI inputs and training.

Read more on this topic

• Munich court rises to prominence in AI copyright litigation
• AI opt-out registry for content creators backed in new study
• New UK bill seeks to address AI copyright issues
• EU AI Act transparency guidelines issued
• The UK AI policy pivot signalled by new copyright reports
• ‘Workable’ AI copyright solutions lacking amidst UK policy ‘reset’

Wyn Davies said: “This case demonstrates the challenge faced by the European Commission and legislators around the world to determine clear and certain standards for allowing AI providers lawfully to use rightsholders’ copyright works.”

“In determining Ross Intelligence’s use of the Westlaw headnotes was not ‘fair use’, the US court considered: the purposes and character of use of the headnotes; the substantiality of the portion used and its reasonableness; and the effect on the market and value of the copyright work, alongside the claimed public benefits. Consideration of these factors in any new standards, legislative or voluntary, will be key to ensuring fairness to unlock the full potential of AI innovation in Europe and beyond,” she said.

Gill Dennis, also of Pinsent Masons, said the significance of the Thomson Reuters decision lies not in establishing a general rule that AI training cannot amount to fair use, but in the court’s close focus on purpose and market substitution.

“Where copyright content is used to train a commercial AI product that competes directly with the rightsholder, particularly where a licensing market for that content is developing, the decision suggests that the economic impact of the use may be central to the fair use analysis,” she said.

“The decision in the case cannot simply be transposed into EU or UK law, not least because the US fair use doctrine differs materially from the EU and UK’s more specific copyright exceptions. Nevertheless, the court’s focus on whether the AI use competes with the rightsholder and affects an emerging licensing market provides a useful lens through which to consider the balance between AI innovation and rightsholder control that EU and UK policymakers are currently seeking to strike,” she said.

Existing EU copyright law applicable to AI training is set out in three different frameworks.

While baseline rights and protections are contained in the 2001 information society directive, EU law provides a broadly applicable text and data mining (TDM) exception that enables AI training unless the rights holder explicitly exercises an opt‑out in a machine‑readable format. Only then does licensing become necessary for the relevant content. The TDM exception and opt-out requirements are provided for in the 2019 digital single market directive. The 2024 AI Act outlines further transparency obligations specific to providers of ‘general purpose AI’ (GPAI) models that extend to copyright matters.

Despite these existing rules, the Commission said content creators and AI developers still face problems.

“Rightholders encounter difficulties to conclude licensing agreements and to get remunerated for their investments in creative content,” the Commission said. “AI providers also face difficulties to access high-quality data stemming from the EU's rich creative ecosystem, which is essential to unlock the full potential of AI innovation in Europe.”

“Particular challenges are faced by SMEs and start-ups that lack resources to navigate the complexities around the use of copyright-protected content for AI. It is essential that these challenges are addressed to make Europe an attractive place for AI innovation and development and deliver the EU's ambition to make Europe an AI Continent with frontier and sovereign AI capabilities, while ensuring that the EU copyright legal framework is balanced, innovation-friendly, and provides legal certainty for both rightholders and AI model and system providers,” it added.

As well as consulting on whether intervention is necessary and, if so, the nature of that intervention, the Commission has invited views on specific mechanisms for supporting enforcement and licensing.

On enforcement, one option under consideration is the development of new standards for opt out protocols and a new EU registry for opt-out. These solutions have long been under consideration across Europe, with UK government ministers stating earlier this year that there remains a lack of workable solutions.

Further options include the introduction of new obligations on AI providers. These could, according to the Commission’s consultation, involve the disclosure of their use of web crawlers to scrape content and record-keeping duties pertaining to their use of copyright-protected content. Another option under consideration is to require AI providers to share details about their use of copyrighted content with rightsholders under non-disclosure agreements.

The Commission also suggested “a rebuttable presumption of use of copyright-protected content by AI providers” could be introduced, subject to conditions – such as where providers do not meet their existing AI Act transparency duties.

On licensing, a range of options were also listed. Under one option, AI providers would be required to pay “fair compensation” when relying on the TDM exception to use copyrighted content.

Other options include the imposition of a new duty on both rightsholders and AI providers to conduct negotiations in good faith, and setting up an independent dedicated dispute resolution forum to determine remuneration AI providers should pay rightsholders to use their content. The Commission could further develop technical solutions to facilitate automated licensing at scale, draw up of model clauses for licences that would address AI use, and promote collective licensing, according to its consultation.

The European Commission’s consultation also invites views on whether EU law needs to be strengthened to “address the use of performers’ likeness in AI-generated content”.

Prominent celebrities including Hollywood actor George Clooney have led calls recently for actors and other performers to be given enhanced rights over use of their voice amidst concern about third parties creating imitations via gen-AI tools.

The European Commission has asked stakeholders whether “EU copyright legislation would be appropriate to address these issues” or whether the issue would be better addressed via “other legal means”.

Pinsent Masons recently highlighted how a variety of different legal frameworks are relevant to ‘digital replicas’ of people’s voices or appearance – and highlighted specific data protection questions businesses wishing to use AI to generate digital replicas for entirely legitimate purposes need to ask themselves first.