Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
In May 2025, more than 400 creative figures, businesses and organisations urged Prime Minister Keir Starmer to require AI developers to disclose which copyrighted works they use to train models. Their letter backed stronger transparency and creator control as Parliament considered a bill amendment—and challenged a government proposal under which copyrighted material could be used for AI training unless rights holders opted out. By March 2026, the government had moved away from that proposal as its preferred option. The campaign helped put transparency at the centre of the debate, but it did not itself establish a complete licensing or disclosure regime.
What the open letter asked for
The letter, dated May 10, 2025, was addressed to Starmer ahead of a key House of Lords vote on the Data (Use and Access) Bill. Its signatories called for the government to enforce copyright law and support a transparency amendment proposed by Baroness Beeban Kidron. The amendment aimed to make AI developers disclose copyrighted works used in training, so rights holders could find out whether their material had been used and consider their options.
The group was described as more than 400 signatories—not exactly 400, and not just individual performers. Alongside artists were creative businesses, media leaders, trade bodies and cultural organisations. The letter and its signatory list show the breadth of the coalition.
Among the best-known names from music were Paul McCartney, Elton John, David Furnish, Dua Lipa, Florence Welch, Kate Bush, Coldplay, Shirley Bassey, Wretch 32 and Robbie Williams. The list also included actors and writers Ian McKellen, Kazuo Ishiguro, Richard Curtis, Russell T Davies, Tom Stoppard and Moira Buffini; visual artists and designers Rachel Whiteread, Antony Gormley, Tom Dixon and John Pawson; and the Royal Shakespeare Company. The campaign was therefore a cross-sector appeal, not solely a musicians’ protest.
#1 Best Overall
Why transparency mattered to creators
Transparency is not the same as banning AI training. It means making information about training material available in a way that helps answer practical questions: Was a songwriter’s catalogue used? Did a publisher’s books enter a dataset? Can a photographer identify images included in training? If so, could the rights holder seek a licence, challenge an unauthorised use or negotiate payment?
The signatories argued that disclosure was necessary for creators to exercise meaningful control and for a licensing market to work. If a creator cannot discover whether a work was used, it is difficult to decide whether to object, seek advice or negotiate permission. The Music Publishers Association’s account of the campaign likewise connects transparency with licensing and rights-holder control.
That does not settle the legal questions. Whether a work was copied, whether permission was required, what a model retains, and whether a particular generated output infringes copyright are related but distinct issues. Training on a work does not, by itself, prove that every later output reproduces protected expression. The letter focused on upstream access to works, disclosure and the ability to license—not a final ruling on every possible AI-generated output.
Free tools Windows power users keep installed
One-click scans. No signup required.
The opt-out proposal behind the dispute
The debate grew out of the UK government’s December 2024 consultation on copyright and AI. One option under discussion would have allowed text-and-data mining of copyrighted material for AI training, subject to a rights-holder opt-out. In plain terms, the difference between the competing models is:
- Opt-in: a developer gets permission before using a protected work.
- Opt-out: use is allowed by default unless the rights holder effectively signals that it is not permitted.
- Transparency: developers disclose what works or sources they used, whether the underlying system is opt-in or opt-out.
- Licensing: rights holders and AI developers agree on permission and, potentially, payment.
Campaigners feared that an opt-out arrangement would shift the practical burden onto individual creators: they would have to discover use, understand how to reserve rights, and ensure that reservations were recognised by developers. That burden could be especially difficult for freelancers and small companies without legal teams or bargaining power. Their concern was not simply that AI exists, but that a default-use system without useful disclosure could leave creators unable to make informed choices or share in value generated from their work.
The government’s stated policy challenge was to balance creator control, transparency and licensing with access to data for AI development. Reporting at the time described ministers’ efforts to revise the proposal amid pressure from the creative industries; see The Guardian’s February 2025 coverage. The letter presented the creators’ position, not a settled legal finding that every use was unlawful. Language such as “theft” or “stolen” should be understood as campaigners’ characterisation unless a specific case establishes infringement.
What happened in the House of Lords?
On May 12, 2025, peers considered the Data (Use and Access) Bill and Kidron’s proposed transparency amendment. The open letter gave the amendment high-profile support just before the debate. According to the Music Publishers Association, peers backed a transparency amendment, while the government opposed the proposed approach and continued to examine alternatives. The Hansard record of the debate documents the parliamentary arguments.
A Lords vote on an amendment is not the same thing as a completed, comprehensive copyright framework. It should not be read as proof that the May proposal, by itself, made every AI developer publish a complete training-data list or created a universal licensing obligation. The distinction matters: an amendment’s progress through Parliament, the final text of a bill and the rules ultimately in force are separate stages.
What would useful disclosure require?
A generic statement that a model used “publicly available data” would tell an artist little. A useful system would need enough detail to identify works or sources, a route for rights holders to search or request information, and a way to check that a developer’s account is reliable. It would also need enforcement; disclosure without consequences for omissions could be hard to rely on.
There are real design tensions. Developers may argue that publishing complete datasets would reveal trade secrets, expose sensitive sourcing arrangements or create security risks. One possible compromise is not necessarily a public dump of every file: the January 2025 Lords debate discussed confidential access and other ways to provide information to regulators or appropriate rights-holder representatives. Options include standardised dataset reports, searchable registries, independent audits or confidential disclosures. Each raises questions about who qualifies for access, how specific the information must be and how disputes are resolved. See the January 28 Hansard debate for discussion of confidential disclosure.
Any system would also have to work for more than major catalogues and publishers. It should be usable by independent musicians, illustrators, photographers, journalists and small production companies, including those whose work is pseudonymous or difficult to identify in a dataset. International reach is another challenge: rules in the UK may not neatly reach every model trained elsewhere, even where that model is offered to UK users.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsThe March 2026 change of direction
The story did not end with the 2025 letter or Lords debate. On March 18, 2026, UK Music reported that Technology Secretary Liz Kendall had said the proposal to let AI companies use copyrighted works without permission was no longer the government’s preferred option. UK Music described this as a move to ditch the plans that had prompted strong opposition from the music sector. Read the organisation’s March 18, 2026 response for its account.
This was a significant policy retreat from the proposal at the centre of the 2025 backlash, not proof that every creator demand had been resolved. The reported change does not, on its own, establish a final system for training-data disclosure, mandatory licensing, compensation or enforcement. Those details—and how any rules would apply to developers operating internationally—remain the substance of the policy question.
What the campaign did, and did not, achieve
The letter made clear that opposition to a default opt-out approach extended well beyond a handful of high-profile musicians. It helped focus parliamentary and public attention on a practical issue: creators cannot effectively exercise rights they cannot see or enforce. The later government change of direction indicates that the original proposal lost support as its preferred route.
But a change in policy preference is not the same as a complete settlement between the technology and creative sectors. The unresolved questions are concrete: what information developers must disclose, who can inspect it, how creators can challenge incomplete records, whether permission or payment is required, and what remedies are available when obligations are breached. The 2025 campaign’s central point remains relevant to each: without dependable information, consent and licensing are difficult to make meaningful.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
