Turning copyright on its head: The UK’s proposed ‘copyright exception’

The UK government’s proposal would grant AI developers a free pass to use the intellectual property of all British writers without their consent and without any payment. AI developers in the United States have been taken to court over alleged copyright violations, and many have signed deals to legally license high-quality copyrighted material to train their AI models. Image by Oskar Młodziński from Pixabay.

By Victoria Strauss

Editor’s note: The UK government’s proposal to hand over the copyrighted work of British writers to AI developers, without permission or payment, has alarmed writers and AI policy experts worldwide.

This article by Victoria Strauss is one of the best overviews of the current UK situation. We’re reposting with the permission of Strauss and Writer Beware, where it originally appeared on Jan. 31, 2025. Writer Beware, co-founded by Strauss, is a publishing watchdog website sponsored by the Science Fiction and Fantasy Writers Association.

The explosive development and spread of generative artificial intelligence (GAI), with its potential for embedding itself in every aspect of our lives, poses a dizzying plethora of issues: environmental, social, existential, even apocalyptic. In this post, I’m going to be focusing on an issue of primary concern to those of us in the creative industries: the struggle to apply the framework of copyright law to GAI companies’ utilization of our work for training their systems.

Developing and training the networks and large language models that power chatbots like ChatGPT and image generation systems like Midjourney requires vast amounts of data scraped from the internet: text, images, music, and more, including material protected by copyright. As we’ve been discovering over the past couple of years, GAI developers haven’t exactly been scrupulous about obtaining the permission of creators for the use of this material.

GAI companies argue that it’s not possible to develop their products without using copyrighted materials, and in any case, that to do so is fair use.* Creators, on the other hand, want to be able to control access to their work, as guaranteed by copyright law–and, incidentally, get paid for the exploitation of it.

There’s enormous anger in creative communities not just over these issues, but over the ways in which GAI can potentially supplant human creators and destroy livelihoods–not to mention the degree to which it is enshittifying the creative sphere by flooding the zone with GAI slop.

But the GAI genie is out of the bottle and will not be put back in. It’s not certain what this will ultimately mean for creators, but it’s certain that it will be part of our creative lives.

ai & copyright: a battle engaged on multiple fronts

Exactly how (or if) creators’ issues of concern will be addressed is being established, minute by minute in real time, by developments on multiple fronts.

Those fronts include:

Legislation, such as the US’s proposed Generative AI Copyright Disclosure Act, which requires GAI companies to disclose their use of copyrighted works (California passed a similar bill in 2024), and the EU’s AI Act, which addresses transparency and creators’ rights.

Litigation. Lots and lots of litigation. Publishers, individual authors, news organizations, musicians, visual artists,YouTube creators, and more have launched lawsuits against various GAI companies for copyright infringement. These legal actions have not always fared well–but they keep coming. In the most recent case, OpenAI is being sued by a group of Indian digital news outlets for the alleged scraping and adapting of their website content.

A growing number of collective licensing schemes. These aim to regularize GAI’s exploitation of copyrighted material, providing authorizations and collecting payments on behalf of rights holders (so many of them, in fact, that there’s now a trade group, the Dataset Providers Alliance). For instance, Calliope Networks, which licenses audiovisual content, and Created by Humans, which licenses books and has partnered with the Authors Guild (itself a vigorous advocate of licensing schemes).

Individual/one-off licensing. For example, last year HarperCollins offered one-time AI licensing deals to many of its nonfiction authors. And a wide variety of media organizations have made individual deals with OpenAI and other GAI companies.

Adapting contract language to address GAI. For the book business, the Authors Guild and others are creating AI-specific contract clauses and insertions, both to protect authors against the employment of GAI in editing and production, and to protect publishers against undisclosed GAI use by authors. I’m seeing such clauses more and more often in the contracts I’m asked to comment on.

A variety of other efforts, including attempts by copyright authorities to interpret how copyright law pertains to GAI (for example, the US Copyright Office has determined that while purely GAI-generated works are not copyrightable, human-authored works that incorporate GAI-generated material are); initiatives such as Fairly Trained, which certifies GAI companies that obtain licenses for the content they use; advocacy groups like the Human Artistry Campaign and the Creative Rights In AI Coalition; and the (largely pointless, IMO) “human certification” programs for writers that have sprung up in the past couple of years.

The uk government’s proposal: a free copyright waiver

This patchwork of responses, however, doesn’t offer a broad-based solution, something that would encompass all creative work and all GAI use of it, and standardize the way it all works. Such a solution is of obvious interest to GAI companies, which want as much access, with as little oversight, as possible–and also to governments wrestling with the thorny issue of how—or if—to regulate GAI.

Now the UK government believes it has come up with just such a solution.

As part of a larger plan to turbocharge GAI development and adoption in Britain, the UK has put forward a series of proposals addressing issues around GAI and copyright–including a significant change to copyright law: the creation of an exception to copyright that would allow GAI companies to train on internet material without having to seek permission from creators. In exchange, GAI companies would be required to offer “increased transparency” on the content they use and where they get it; and creators would have the option to “reserve their rights”–i.e., to opt out.**

deadline for public comment is midnight, Feb. 25

Comments from the public are currently being sought via an Open Consultation (you can see the government’s overview, as well as the questions it is asking the public to weigh in on, here).

Quoting the government’s post:

This consultation seeks views on how we can deliver a solution that achieves our key objectives for the AI sector and creative industries. These objectives are:

1. Supporting right holders’ control of their content and ability to be remunerated for its use.

2. Supporting the development of world-leading AI models in the UK by ensuring wide and lawful access to high-quality data.

3. Promoting greater trust and transparency between the sectors….

The proposals include a mechanism for right holders to reserve their rights, enabling them to license and be paid for the use of their work in AI training. Alongside this, we propose an exception to support use at scale of a wide range of material by AI developers where rights have not been reserved….

The government believes that the best way to achieve these objectives is through a package of interventions that can balance the needs of the two sectors [creators and GAI companies]. That is why we are consulting on measures that would require increased transparency from AI developers. This includes the content they use to train their models, how they acquire it, and any content generated by their models. And it is why we are consulting on the introduction of an exception to copyright law for “text and data mining”. This improves access to content by AI developers, allowing right holders to reserve their rights and thereby prevent their content being used for AI training. Progressed together, we believe these measures could meet our objectives above.

this turns copyright on its head

But what the UK is proposing isn’t just an “exception” to copyright law. It turns copyright on its head.

Copyright is an opt-in system. As the copyright owner, you control access to your work, which can be exploited by others, including for commercial purposes, only with your permission. Removing the permission requirement, with your only remedy being to say “don’t do that”, is exactly backwards.

Additionally, opt-out places the burden of protecting creative work entirely on the creator–a huge and possibly unachievable task, as this essay from Fairly Trained founder Ed Newton-Rex points out. “It is impossible to opt out downstream copies of your work”, he writes, simply because there’s no opt-out scheme that is capable of catching every use.

Quoting from Newton-Rex’s essay:

The most widely used opt-out schemes are location-based: for instance, robots.txt, which contains information about how the web domain where it’s hosted can be accessed by web crawlers. These only work for web domains you control. But your work is likely to appear in many other places: for instance, a photographer’s photo may feature in an ad on another website; a composer’s song may be used in an online video; screenshots of a newspaper article may be shared on social media. You have no control over whether these downstream uses of your work are opted out of generative AI training, since you do not control the URLs where they’re hosted.

Unit-based opt-out schemes, on the other hand, add metadata to content that indicate the content’s opt-out status. But metadata is easily removed (e.g. X removes EXIF metadata when images are uploaded) and, besides, there is no way to add metadata to text. Unit-based opt-out schemes fail to opt out downstream copies where metadata cannot exist in the first place or does exist but is subsequently removed, which are common.

There are no effective opt-out schemes that reliably opt out content itself from training. Doing so is impossible. One theoretical solution could be maintaining a central directory of opted-out works and using automatic content recognition (ACR) to scan datasets for opted-out content; but even this does not work, as ACR cannot reliably identify opted-out works. Relying on ACR for opt-out information would allow modified versions of opted-out content to be trained on, as well as individual copyrighted elements that make up part of a larger copyrighted work (e.g. the composition in a sound recording, the lyrics in a song, or the dialogue in a film), bootlegged recordings of live performances, and works transposed from one medium to another (e.g. screenshots of text). None of these would be caught by automatic content recognition systems scanning for the opted-out works.

Could future innovation create a reliable opt-out system? Possibly—but even if it could, saddling creators with the enormous administrative responsibility of policing all conceivable iterations of their work “is evidently unfair”. (The essay, which discusses multiple other reasons opt-out doesn’t work, is worth reading in its entirety.)

UK writers and artists: raising the alarm

The creative community has reacted to the UK’s package of proposals with alarm. Individual creators are voicing opposition. Industry groups have raised concerns. Organizations representing creators are urging their members to comment during the Open Consultation and to reach out to their MPs, as an opportunity to weigh in on, and potentially drive the direction of, UK law. So far, more than 40,000 people have signed the Creative Rights in AI Coalition’s Statement on AI Training, which calls for the end of unlicensed use of creative works by GAI companies.

It’s much too soon to know whether the proposals will become law–but even if they don’t, similar schemes will surely be proposed in other countries eager to reap the economic benefits of GAI–so even if you’re not a UK citizen, this is an issue you should care about. The UK’s proposed copyright exception is a dangerous change that, under the guise of expanding creators’ ability to protect their work, actually diminishes their rights–a glimpse of a GAI future in which creators always come second, and must always be the ones to accept compromise.

The Open Consultation will be live until February 25. UK citizens can comment here. The Society of Authors is also seeking comment; you don’t have to be a member to fill out its survey, which you can find here.

* There are echoes here of an earlier fight over fair use. In 2005, the Authors Guild and a group of publishers sued Google over its mass digitization of in-copyright print books, arguing that Google had not sought creators’ permission and therefore was committing copyright infringement. Google argued that the digitization was fair use. The courts ultimately sided with Google, and a subsequent ruling in a separate book scanning case confirmed the fair use argument.

** The opt-out question also echoes in the Authors Guild’s Google lawsuit. In an effort to settle the suit, a proposed compromise was worked out in which Google could proceed with its book scanning project, with authors having the remedy of opting out if they didn’t want their books included. Multiple writers’ groups opposed the settlement, with the inversion of copyright law being a major factor. The settlement was ultimately denied, sending Google and the AG back to the drawing board.