AI Regulation Tracker / Policy decision
UK Government Drops Its Broad AI Text-and-Data-Mining Copyright Plan
In its statutory Report on Copyright and Artificial Intelligence, published March 18, 2026, the UK government abandoned its previously preferred plan for a broad text-and-data-mining exception with a rights-holder opt-out. This is a policy decision, not a new law. The existing narrow exception for non-commercial research stays in place.
On March 18, 2026, the UK government published its Report on Copyright and Artificial Intelligence. This was not a discretionary blog post from a ministry. It was a statutory report, one the government was required to produce under section 136 of the Data (Use and Access) Act 2025, which passed in 2025 and obliged ministers to report back on the copyright and AI question after a long and contested consultation. The consultation drew 11,520 responses, which tells you how loud the fight over training data had become in the UK.
What the government actually decided
The headline is a reversal. Earlier in the process, the government had signaled that its preferred route was a broad exception to copyright for text and data mining, coupled with a mechanism that would let rights holders opt out of having their works used. In the consultation this was Option 3, and it was the one ministers leaned toward. The idea was to give AI developers a wide legal path to train on copyrighted material unless a rights holder actively said no.
In this report, the government walked that back. The line that matters, in the report's own words, is this: "A broad copyright exception with opt-out is no longer the government's preferred way forward." That is a change of position, stated plainly. The route the government had favored is off the table as the favored route.
It is worth being careful about what that sentence does and does not mean. It means the government has stopped preferring the broad exception with opt-out. It does not mean a law was passed, and it does not mean anything was repealed. There was no exception on the books to repeal here. What changed is the government's stated direction of travel, not the statute.
The existing narrow exception is untouched
The UK already has a text-and-data-mining exception in its copyright law, but it is narrow. It covers non-commercial research, and it has for years. That is the exception a university researcher relies on when mining text for a study that is not being run for profit.
That narrow, non-commercial-research exception stays exactly where it was. The report did not expand it into commercial territory, and it did not take it away. So the practical position in the UK today is the same as before on the law itself: non-commercial research mining is permitted under the long-standing exception, and there is no broad commercial training exception. What is new is only that the government has dropped its plan to create one along the opt-out model it had preferred.
Transparency: work with industry, not legislate now
The other live issue in this debate has been transparency, meaning whether AI developers should be required to disclose what copyrighted material they train on. Rights holders have pushed hard for a legal transparency duty, arguing that they cannot enforce their rights if they cannot see what is being ingested.
On this, the government chose the softer path for now. Rather than legislating a transparency requirement in this report, it said it will work with industry on best practice. That is a decision to hold off on a statutory mandate and pursue voluntary or collaborative standards instead. It is not a permanent no to transparency legislation, but it is a clear signal that the government is not reaching for a legal duty at this stage.
What this is, and what it is not
I want to be precise, because a story like this gets flattened fast.
This is a policy decision, published in a report the government was legally required to write. It is a statement of position and direction. It is not a new copyright law, not a new exception, and not the repeal of an old one. No AI developer gained a new legal right to train on commercial works on March 18, and no rights holder gained a new enforceable claim. The legal architecture of UK copyright, including that narrow non-commercial-research exception, is the same today as it was the day before the report came out.
What moved is the government's plan. It had a preferred solution, the broad exception with opt-out, and it has now said that is no longer its preferred way forward. Anyone who reads that as "the UK banned AI training" or "the UK legalized AI training" has misread it. Neither happened. The government stepped back from a specific legislative design and pointed at further work.
Why US AI labs and finance leaders should care
The UK is not a US regulator, and nothing in this report reaches a US filing or a US company's obligations. But the signal travels, and for US AI labs it is a material one.
The broad exception with opt-out was, in effect, the friendliest realistic outcome for model developers who want legal certainty to train on copyrighted works at scale. It would have given them a wide default permission. By dropping it as the preferred route, the UK is telling the market that it will not, for now, grant a commercial training exception. That pushes the practical answer back toward licensing. If you want UK-copyrighted works in your training set with legal comfort, the path the government is leaving open looks a lot more like paying for it than like relying on a broad statutory carve-out.
For finance and strategy leaders at AI companies and their investors, that is a cost-and-risk input. Training-data licensing is a real line item and a real liability question, and a major English-speaking jurisdiction just declined to make it cheaper by statute. It also lands alongside how other jurisdictions are wrestling with the same tension, and the direction here is toward negotiated licensing and away from a broad free pass. Treat it as a planning signal on where training-data economics are heading, not as a rule that binds anything you file in the US.
What to do now
Read the actual decision, not the headline. The government dropped its preferred broad exception with opt-out; it did not pass or repeal a law, and the narrow non-commercial-research exception still stands. If you run or advise an AI developer touching UK-copyrighted works, assume the near-term route to certainty is licensing rather than a broad exception, and budget for it. Watch the transparency track, because the government chose best-practice collaboration over legislation for now, and that could shift. And do not let anyone on your team describe this as the UK enacting or banning an AI copyright exception. It did neither. If you make a real decision on training data, base it on legal advice about the actual state of UK law, not on a summary of a policy report.
Questions professionals are asking
Did the UK pass a new AI copyright law?
No. This is a statutory report, required under section 136 of the Data (Use and Access) Act 2025 and published March 18, 2026. It states a change in the government's position. It does not enact, amend, or repeal any copyright exception or legal duty.
What exactly did the government drop?
Its previously preferred plan for a broad text-and-data-mining exception with a rights-holder opt-out, known in the consultation as Option 3. The report says a broad copyright exception with opt-out is no longer the government's preferred way forward.
Does the UK still have any text-and-data-mining exception?
Yes. The existing narrow exception for non-commercial research remains in place and was not changed. What the government abandoned was its plan to create a broader, commercial-facing exception on the opt-out model. There is still no broad commercial training exception.
What did the report say about transparency?
The government decided to work with industry on best practice for now rather than legislate a transparency requirement. That is a decision to hold off on a statutory mandate at this stage, not a permanent ruling out of transparency rules later.
Why does this matter for US AI labs?
The UK is not a US regulator, so this binds nothing in the US. But by declining, for now, to grant a broad commercial training exception, a major English-speaking jurisdiction is signaling that the realistic near-term path to certainty for training on UK-copyrighted works is licensing. That is a material input for US labs weighing training-data cost and legal risk.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any law, exception, or policy applies to your situation with qualified professionals in the relevant jurisdiction.