After months of intense lobbying from the UK's creative industries, the government has confirmed it is no longer pursuing a broad copyright exception for AI training — a policy that would have allowed technology companies to use copyrighted works freely as training data without seeking permission or paying rights holders. The reversal marks one of the most significant policy shifts in the ongoing global debate over AI and intellectual property.

The Publishers Association, the Society of Authors, the Authors' Licensing and Collecting Society (ALCS), and individual publishers including Pan Macmillan all welcomed the announcement. Sara Lloyd, CEO of Pan Macmillan, described the decision as "a positive step towards ensuring that human creativity is valued and protected in the age of AI" — a formulation that captures the industry's cautious but genuine relief.

The UK government had previously indicated a preference for a broad text-and-data mining exception modelled loosely on approaches taken in Japan and Singapore, where AI training on publicly accessible content is permitted without licence. That position drew fierce opposition from the creative sector, which argued it would effectively allow AI companies to extract value from decades of human creative labour without compensation, undermining both the economics of publishing and the incentive structures that sustain original authorship.

The government's revised position does not yet constitute settled law. Officials have indicated they will continue to consult with both the creative industries and the AI sector before introducing any legislative changes, and have committed to ensuring that any reforms are "proportionate, evidence-based and aligned with the UK's long-term strategic objectives." The ALPSP, which represents learned and professional society publishers, welcomed the report's emphasis on licensing market development and transparency, noting that a market-led approach can progress provided the government maintains oversight to prevent asymmetries of power between large AI developers and individual rights holders.

For book publishers, the practical stakes are considerable. AI companies have been training large language models on vast corpora of text, including books, without systematic licensing arrangements. The question of whether such training constitutes copyright infringement — and whether a statutory exception should legalise it — is being litigated simultaneously in the United States, where Merriam-Webster and Encyclopedia Britannica filed suit against OpenAI just two days ago. The UK's decision to step back from a broad exception does not resolve these questions, but it does signal that the government is unwilling to pre-emptively legislate in favour of AI companies at the expense of rights holders.

The decision also arrives at a moment when the voluntary licensing market is beginning to mature. Several major publishers have signed content licensing agreements with AI developers, and collective licensing bodies are exploring frameworks for systematic remuneration. The government's recognition that "the evolving licensing market demonstrates that creative materials can add value in the AI supply chain" suggests a preference for market-led solutions over statutory intervention — a position that suits publishers who have already begun negotiating from a position of relative strength.

What remains unresolved is the question of transparency: rights holders still have limited visibility into whether and how their works have been used in AI training. The government has committed to developing best practices on input transparency, but without enforceable disclosure requirements, publishers and authors will continue to operate in a state of informational asymmetry relative to the AI companies that have trained on their content.

**Editorial commentary:** The UK government's retreat from a broad AI copyright exception is a significant moment, but it should be read as a pause rather than a resolution. The creative industries have won a battle, not the war. The fundamental tension between AI developers' appetite for training data and rights holders' legitimate interest in controlling and monetising their work remains unresolved, and the government's commitment to further consultation means the legislative landscape will remain uncertain for the foreseeable future. What the publishing industry now needs is not just the absence of a bad policy, but the presence of a good one: enforceable transparency requirements, a functioning collective licensing infrastructure, and clear remedies for infringement. The voluntary licensing market is a start, but it favours large publishers with the legal resources to negotiate directly. Smaller presses, individual authors, and academic societies remain exposed. The next phase of this debate will determine whether the UK's creative economy emerges from the AI transition with its economic foundations intact.