That door is shut. In its formal response to the House of Lords Communications and Digital Committee’s AI and intellectual property inquiry, the UK government confirmed it won’t introduce a broad commercial text and data mining exception that would allow AI training on protected content without rightsholder opt-out rights. The policy it’s pursuing instead centers on licensing, best practices for AI developers, and voluntary transparency – not a statutory exception.
This isn’t a surprise in direction. The licensing-first signal has been consistent from UK government communications for months. What changed on May 16 is that it’s now formal policy, not positioning. Developers who factored a broad TDM exception into their training data strategy need to update that assumption.
What the decision does and doesn’t require
“Licensing-first” is the editorial characterization of the government’s direction, per the parliamentary response – it isn’t a term the government document uses, and it doesn’t yet mean a mandated licensing scheme with enforcement teeth. What the government response establishes is that the opt-out model championed by publishers and the creative industries will shape the policy framework rather than the broad exception model that AI developers had advocated. Voluntary transparency and developer best practices are the current instruments. Mandatory licensing rules aren’t in the parliamentary response.
UK AI Copyright Policy Positions
Don’t expect that to stay static. The UK government has said what it won’t do. What it will do – specific licensing framework design, standards development, oversight mechanisms – remains under development. The practical consequence now is narrower: AI developers using UK-origin training data no longer have a credible path to a statutory exception covering that use. They’re in a licensing environment, whether or not the formal licensing infrastructure exists yet.
Who this affects and how
The exposure question isn’t abstract. Books, journalism, music, visual arts – UK rightsholders in each category have been pressing for the licensing model the government has now signaled. Developers building or fine-tuning models on corpora that include UK-origin content in those categories should be auditing that exposure now. “UK-origin” in a global training dataset isn’t always visible without deliberate provenance tracking, and that’s exactly the gap a licensing regime will eventually require closing.
The UK decision extends a pattern visible across jurisdictions this month. The US Supreme Court’s denial of cert on AI copyright review last week preserved the human authorship standard. The EU’s approach through the Digital Omnibus keeps GPAI transparency obligations intact. As we mapped last week in our global copyright convergence analysis, the cross-jurisdictional direction is toward greater rightsholder protection, not less.
What to Watch
What to watch
Two things. First, the UK government’s follow-on work on licensing framework design – the parliamentary response signals direction, but the substantive policy mechanism hasn’t been built yet. When that framework surfaces, it will set the terms developers actually have to operate under. Second, the Intellectual Property Office’s best practices guidance for AI developers, which the government response references as a near-term output. That’s where the operational requirements will first appear.
TJS synthesis
The UK’s formal position does something strategically important for rightsholders: it removes ambiguity. A licensing-first environment without a built licensing framework is still an environment where developers face liability exposure for un-licensed use of UK-origin content. Compliance teams at AI developers with meaningful UK-origin training data in their pipelines should treat this decision as the trigger for a provenance audit, not a reason to wait for the licensing rules to formalize. The rules that come next will be easier to comply with if you already know what’s in your training set.