Text and data mining exception

In short

A text and data mining exception is a rule in copyright law that lets people copy works so a computer can analyse them, without the owner's permission. In the UK it only covers non-commercial research, so it does not cover most commercial AI training.

Also called: TDM exception, Data mining exception, Section 29A, Text and data analysis exception
Topic: Law, policy and the UK
Level: Some background helps
First described: Section 29A was inserted into the Copyright, Designs and Patents Act 1988 from 1 June 2014 by the Copyright and Rights in Performances (Research, Education, Libraries and Archives) Regulations 2014 (S.I. 2014/1372)¹.

Text and data mining (TDM) means using software to analyse large amounts of text, images or data to find patterns. Training an AI model involves copying huge numbers of works, so whether an exception applies matters a great deal to AI developers and to creators.

The UK exception is section 29A of the Copyright, Designs and Patents Act 1988, added in 2014¹. It allows someone with lawful access to a work to copy it for computational analysis "for the sole purpose of research for a non-commercial purpose", with an acknowledgement where practical¹. Contract terms that try to block this copying are unenforceable¹.

The EU goes further. Article 4 of its 2019 Digital Single Market Copyright Directive allows mining for any purpose unless the right holder reserves their rights, which in the EU should be done in a machine-readable way². The UK government's consultation noted that the UK exception, unlike the EU's, does not cover commercial research².

In December 2024 the UK government consulted on a new exception for commercial mining with an opt-out for right holders, its preferred option at the time². The Data (Use and Access) Act 2025 then required an economic impact assessment and report³. In the report, published on 18 March 2026, the government said a broad exception with opt-out was "no longer the government's preferred way forward" and that it would not change copyright law until confident the reforms met its aims⁴.

So, as of October 2026, UK law has no exception covering commercial AI training on copyrighted work⁴.

Example

A university researcher with a library subscription can copy thousands of journal articles to study how scientific language has changed, under section 29A¹. An AI company copying the same articles to train a commercial chatbot cannot rely on that exception⁴.

Why it matters

It sits at the centre of the UK fight between AI companies, which want to train on published work, and writers, musicians and artists, who want permission and payment.

Questions people ask

Can AI companies train on copyrighted work in the UK?

Not under a general exception. The UK's text and data mining exception covers only non-commercial research¹. The government reported in March 2026 that it had dropped a broad exception with opt-out as its preferred option and would not change the law until confident in reforms⁴.

What is the difference between the UK and EU text and data mining exceptions?

The EU's Article 4 exception allows mining for any purpose unless the right holder opts out, ideally in machine-readable form. The UK's section 29A covers only non-commercial research and has no opt-out²¹.

Can a contract stop text and data mining for research in the UK?

No. Section 29A says any contract term that tries to prevent or restrict copying allowed by the section is unenforceable¹.

Related terms

Sources

Back to the glossary