Categories
Advice and Guidance

Guidance on resisting restrictive AI clauses in licences

Over the last six months Jisc Licensing has been negotiating with publishers and providers of online content seeking to include clauses in our licences that restrict the use of their content in artificial intelligence (AI) systems. Jisc believes that AI, when used responsibly, can be a legitimate tool to enable learning, teaching and research, which should not be prohibited. We want to support our members in their own dealings with publishers, as a unified position across the sector will strengthen our own negotiations, so have compiled this guidance to encourage you to push back on these clauses in your own licences. It is intended as guidance only and should not be relied on as legal advice.

Jisc’s licences

The Jisc model licence does not include restrictions in connection with the use of licensed content or data (including any associated metadata) for machine learning and / or AI purposes for two principal reasons:

  • Such clauses are both legally complex and challenging to enforce. Aside from the obvious difficulties tracking the use of licensed content in AI (eg for training large language models), some AI systems can automatically generate and manipulate content without direct human intervention. Consequently, the inclusion of a restriction would create significant legal risk for licensees as, practically, they will not be able to track the use of licensed content in this way.
  • Use of licensed content in AI may, in some circumstances, equate to fair dealing where it falls within the text and data mining (“TDM”) exception for non-commercial teaching and research by lawful users as set out under section 29A of the Copyright, Designs and Patents Act 1988 (“CDPA”). Contractual terms in licence agreements do not override the exceptions to copyright infringement and, so long as licensees’ activity meets the criteria for the TDM exception, they will likely fall within the exception.

Pending new policy or judicial clarification on the rules for copyright and AI, Jisc’s view is that the existing terms in our model licence are suitable to manage each party’s rights alongside established statutory rules, and along with other consortia worldwide, we endorse the ICOLC Statement on AI in Licensing.

Your licences

Jisc advises our members to oppose licence clauses that:

  • prohibit the use of AI for legitimate research or educational purposes.
  • restrict usage of licensed content in ways that are fundamentally unenforceable.
  • prevent you from making licensed content fully accessible to all authorised users in any legal manner.
  • introduce new liability on your institution, or require you to indemnify the publisher, especially in relation to the actions of your authorised users.

Making the case

Existing terms are enough

As a first step try to reject any AI clauses altogether on the grounds that existing licence terms and UK copyright law should cover all authorised and unauthorised uses of the licensed content. This will include restrictions on unattributed use of extracts or parts of the content, on making it available to anyone who does not have authorised access, and on excessive or systematic copying. The supplier will probably have security measures in place on its platform that detect and/or prevent excessive use and downloading, which should be applicable to AI use as well. More guidance on copyright law can be found here: An introduction to copyright law and practice in education, and the concerns arising in the context of GenerativeAI.

Unenforceability

Make the case that, increasingly, commonly used applications (such as PDF readers, Microsoft products, browsers, and browser extensions) have embedded AI functionality in resources that interact  with licensed content. It is unfeasible to expect such usage to be monitored and restricted at an institutional level, putting you at unacceptable risk of breaching the licence. It will support your argument if your institution has publicly available library regulations or guidance for staff and students (or better still an institutional policy) that are robust in regard to AI and specify that copyright law and licence terms should be respected.

The TDM exception

Section 29A of CPDA permits anyone who has lawful access to licensed content to make copies of it in order to carry out computational analysis for the purpose of non-commercial research, as long as those copies are not shared with any unauthorised users. Using copied licensed content in generative AI systems can be interpreted as permissible computational analysis, but such use in systems that retain the copy could be interpreted as making it accessible to unauthorised users, which would infringe copyright and so should be avoided and warned against in guidance to users.

The benefits of working with AI

There are aspects of authorised AI use that could benefit publishers and vendors. It could increase the visibility and discoverability of their content and increase their accessibility compliance. It could show that they are adapting to evolving trends in scholarly communication and demonstrate their forward thinking nature. As use of AI in research and education develops, customers’ purchasing decisions may hinge on the levels of support for AI initiatives that vendors offer. All this could lead to new opportunities for collaboration between content owners and content users.

Alternative wording

If the publisher/vendor does not accept these arguments, propose the use of alternatively worded clauses that address the concerns of the publisher while protecting the legitimate uses of AI detailed above. Reinforce the claim that all uses of AI should not be restricted, only the knowing acts that could infringe copyright or the existing licence terms.

 

 

By Ben Taplin

Contracts Specialist · Licensing
Jisc

Leave a Reply

Your email address will not be published. Required fields are marked *