24.01.2025: In a packed room in Westminster, six All-Party Parliamentary Groups (APPGs) came together to hear from industry leaders, including UK Music Chief Executive Tom Kiehl, who expressed concerns regarding the Government’s AI and copyright consultation. 

For the first time since their inception, the APPG on Music collaborated with the APPG for Intellectual Property (IP), APPG for Writers, APPG for Publishing, APPG for Media, and APPG for Visual Arts and Artists for a special, urgent meeting focused on the impact of artificial intelligence (AI) on copyright.

Chaired by former Government Minister the Rt Hon Sir John Whittingdale OBE MP, the event on Wednesday gathered several Parliamentarians to discuss a critical issue that could reshape the future of the UK’s creative industries. 

Speaking to MPs and Peers in attendance, Tom Kiehl outlined the practical challenges of the proposed “opt-out” mechanism, particularly in light of lessons learned at the EU level.  

Highlighting the potential risks to the music industry, he said: 

“Thank you very much for the opportunity to talk to you about some of the concerns from the music industry regarding the proposals by the government for an exception to copyright for text and data mining purposes.  

“In their recently published consultation Government expressed their preference for an exception for text and data mining with a rights reservation, which is more commonly referred to as an opt out. 

“The theory is that a rights reservation mechanism is a safeguard for creators and rightsholders so that they are able to opt out of having their works and creativity ingested by Artificial Intelligence systems and models. I would like to highlight why an opt out does not provide reassurance to music. 

“There are many practical problems with an opt-out mechanism, and these challenges have been evidenced by the experience in the EU where such a provision exist. 

“From a creator and rights holder perspective the main concerns are the operation of a rights reservation in practice and the lack of transparency in particular if the rights reservation has been respected. 

“For an opt out to be truly effective a creator or rightsholder would want to opt out of any AI platform using their work. It is however unreasonable to expect individuals to have to notify all the hundreds and thousands of AI service providers in order to do this. 

“Related to this, many artists and composers will want to be able to opt out their entire repertoire, but it would be unreasonable to expect them to monitor what has happened to their works, especially if it has been transferred to other rightsholders. 

“In the EU, their text and data mining exception was published in 2019, which was before the widespread use of generative AI. Fundamentally, it was not designed with large-scale ingestion of internet content by AI systems in mind. To therefore draw parallels with what exists in the EU, as the UK government is proposing, is out of step with the new reality. 

“The EU experience has also demonstrated a lack of clarity about what exactly constitutes an “opt-out” in the EU. This has led to legal uncertainty. One German court found that a statement in the website’s terms and conditions constitutes an opt-out, though this does not act as a technical block to accessing content and this may not hold up in other courts either. 

“Presently, the most common way of opting out is through the website’s robots.txt file. This is a text file that tells search engines which URLs and directories on a website should not be crawled. 

“Robots.txt and similar opt-outs are unworkable for music, however. Robots.txt is location-based, meaning they only work on web domains controlled by the rightsholder. However, music is typically copied and shared to many platforms beyond the rightsholders control, for example where an artist’s track is embedded within a YouTube video. Without control of the web domain, they are unable to opt-out. 

“Take up from rightsholders to opt-out remains low. This is evidenced by a lack of knowledge about opt out amongst rightsholders, failures to opt-out successfully, or simply an increasing view that the burden is too high to opt-out effectively. For smaller rightsholders this is likely to be particularly the case. 

“Rights reservations in other areas, such as privacy, are also often ignored. A US study on data privacy preferences found 88% of companies failed to respect user opt-out preferences. If companies fail to respect opt-out preferences in data privacy, why would we expect AI developers to honour these opt-outs for large-scale generative AI training? There is also no way of knowing whether a reservation has been respected at present, making it even more likely that AI developers will simply ignore rightsholder requests. 

“And from a technological perspective, if someone was to seek an opt out of a specific piece of music that they had created, this is technically not possible as AI models are not able to selectively remove individual pieces of training data from datasets. Each element of training data updates a huge matrix of probability values and trying to remove the contribution of a single item of data simply doesn’t work. 

“This is even more challenging in situations where training has already taken place. Many creators are therefore left with no control of their works once they have already been ingested by AI systems. 

“Taking this into consideration, effective removal of training data requires a model to be re-trained with the opted-out material removed. Again, this is problematic as it will require huge computer costs to achieve, literally megawatts and gigawatts of energy. It would therefore be hugely costly and inefficient for AI service providers to act upon opt outs. 

“If the Government’s proposals became a reality there would be a need for a centralised, up-to-date depository where AI developers can verify if rightsholders have opted out of having their work used for AI training. However, this does not exist. 

“In short, proposals for a rights reservation, or opt-out, are deeply flawed and provide no reassurance whatsoever to a sector like music. AI has many benefits to music, from enhancing the creative process to improving business efficiencies, yet proposals for a text and data mining exception would create huge damage to the sector that contributes £7.6 billion to the economy, generates exports of £4.6 billion and employs 216,000 people”. 

The meeting saw insightful questions from MPs and Peers about the practical next steps, including the possibility of a Westminster Hall debate and strategies for encouraging more members of the public to submit responses to the consultation. 

A number of Parliamentarians joined in person including Rt Hon Sir John Whittingdale OBE MP, Pete Wishart MP, Patrick Hurley MP, Sharon Hodgson MP, James Frith MP, Dawn Butler MP, Polly Billington MP, Baroness Kidron, Lord Clement-Jones, Lord Foster of Bath, Earl of Devon, Lord Stevenson, Lord Black, Baroness Bowles, Earl of Clancarty, Lord Watts and Lord Freyberg.  

In addition to the Parliamentarians in the room, over 70 attendees joined the discussion online, representing a wide spectrum of cultural and creative industries.