With all the buzz about how AI will change our lives, the controversy over changes to copyright law and the potential for new revenue streams from licensing content to AI developers, how much thought has been given to the impact on publishers’ permissions activity? Probably not very much.
There are some important implications for those involved in granting permission to re-use extracts of content, and workflows will need to adapt accordingly.
This blog covers some of the questions that we think permissions teams will probably need to be asking when dealing with permissions requests in the age of AI, if they aren’t already.
Typical permissions questions reconsidered in the age of AI
Did we publish it (like that)?
When you receive a permissions request, the first question asked is usually: “Did we publish it?”
I think that question changes slightly in the age of AI, or at least there is a supplementary question: “Is the content requested what we actually published?”
If requestors have used AI for part of their research or writing work, it’s possible that any outputs received have been ‘hallucinated’. That is that the AI has perceived patterns in data that don’t really exist, resulting in outputs that are inaccurate or in some cases nonsensical. You could face a situation where you are being asked for permission to reproduce something that the requestor believes to be an accurate reproduction (based on their AI research), when it is not.
It is also possible that incorrect source information has been supplied and requests may as a result, be misdirected. It is worth verifying then that the content being requested is both something that you published and is being reproduced accurately.
Do we have the rights?
The second question a permissions professional will ask is ‘do we have the right to grant permission?’ a question that becomes even harder to answer when you factor in requests which ask you to include the right to use the content in any subsequent AI licensing deals. Therefore it is important to consider whether or not the agreements you have in place for the requested content, allow you to grant the specific rights requested. Given the range of publishing agreements that may have been entered into by a publisher over time, the question of whether a particular work is OK for AI licensing or not, is not straightforward. Even where your agreements are clear that you have the rights to grant, they may require you to include additional restrictions or obligations to convey to the requestor.
Are the author’s moral rights impacted?
We don’t often talk about moral rights but in the UK and Europe it is particularly important to consider these when granting permission. Even if moral rights don’t apply in your jurisdiction or you have moral rights waivers in your agreements with creators, we would suggest it’s still good practice to consider the impact of the re-use rights you grant on your creators and to try to mitigate any potential negatives. Will the creator be identified as the author of the work when it is reproduced in the requestor’s product? (See below for more on acknowledgements.) Have there been any adaptations or changes made to the original (as above this may well be the case if the content requested has been generated via AI)? Changes and adaptations, even quite subtle ones could be damaging to an author’s reputation, so you might want to check that your author is happy for the content to be reproduced as requested before granting permission. Under UK legislation one right authors have is not to have work falsely attributed to them. The risk of this happening is higher when using AI, where content is created from multiple sources and hallucinations are possible. This underlines the importance of verifying that the content is exactly as per your author’s original published version.
Just for fun I asked Gemini (Google’s generative AI tool) to give me a quote that demonstrates the power of writing as a thinking tool from my colleague Alison Jones’s book Exploratory Writing.
It was unable to give me a quote as it didn’t have access to the full text (good news!) but to get around that it gave me a paraphrased quote gleaned from reviews, blurbs etc., that it assured me “accurately represented her view of writing as a tool for thinking”. Nice to know it is so confident! It is easy to see from this example, how writers using AI tools could end up getting versions of content that are not quite what they thought they were getting, especially if they are not meticulous about recording source data, or don’t read the small print about what they are being served.
What terms and conditions should apply?
It is worth looking at your standard terms and conditions to determine if changes are required to accommodate AI use. It would be worth discussing and reviewing your T&C with legal counsel and colleagues from across your organisation, and making sure that you are applying T&C to every request, even ones you grant free of charge.
We now live in a world where AI is being used by everyone, all the time, often without the people using it even being aware. Even if the requestor is not specifically asking for the right to include your content in any AI training licences, chances are at some point the ‘downstream’ copy of the content you are granting permission for is likely to be crawled or scraped by AI software. Print documents can be scanned and uploaded by those with physical copies, which may then be used in a generative AI query and go on to form part of the training data for the underlying model. Neither you nor the entity you granted the permission to has any knowledge of or control over this, and often users of generative AI software have not read the small print that says that data they upload in prompts may be used as training data. This means they too may be unaware that their use of such tools without rights holder permission would be an infringement of copyright.
As AI usage becomes more ubiquitous, how do you restrict AI access to content that you don’t have the rights to, or would otherwise prefer not to be scraped?
It’s a difficult question to answer and at the moment options are limited, none are guaranteed to prevent your content ending up in training data.
One suggestion is to really amp up your acknowledgement requirements so that the origin of the content as well as the rights holder with responsibility for licensing the content is clear and that the acknowledgement is placed in close proximity to the content. If you don’t want your content to be scraped, as part of the acknowledgement you could ask the requestor to add a notice explicitly stating that you are reserving all rights and that the material may not be used for AI training (talk to your legal counsel for advice on the most appropriate form of words). It remains to be seen where the proposals to amend copyright law and the various copyright infringement cases being brought against AI companies will take us, but a clear rights reservation statement next to your IP is as strong a message as we currently have available to us and may be important in the event of future enforcement action. Ideally we would have a technological means of identifying copyright content anywhere it appears online and be able to define what is or is not permitted in terms of re-use that both humans and bots could understand, but we are a long way off that yet.
Are requestor’s compliant?
You can put whatever you like in your terms and conditions but unless you check what people are ACTUALLY doing with the content you’ve granted permission for, then you may find that your T&C are not being complied with.
As a minimum, we’d suggest checking that acknowledgements and or reservation of rights statements are present, correctly located and have been reproduced accurately. You could also check that the content itself has been reproduced accurately and that nothing in the editing/proofing stage (which may have been carried out using AI tools) has introduced errors or changes that have not had prior approval. Going back to a non-compliant requestor to remind them of the terms and conditions of use and asking them to rectify (pointing out that their permission in invalid unless they do) is usually enough to bring things back in line with your T&C.
Do you have more questions about granting permissions in the age of AI?
These are just are some of the things that we’ve been thinking about when it comes to granting permissions. If you work in granting permissions, what questions have you been asking?
If you have any more to add to our list, we’d love to hear them. Email info@rightszone.co.uk
IMAGE CREDIT: @billion-images