AI, authorship, and risk management

On 31 July, it was announced that a high-profile publishing deal reportedly worth £1.5m for a debut crime novel had collapsed following concerns that the author had utilised artificial intelligence (“AI”) when writing it.

Minotaur, an imprint of Macmillan US, had plans to publish the works in 2028 but withdrew their offer following receipt of an email from the author’s agents, in which the agents explained that they were “no longer able to authenticate how the manuscript fully evolved from origin to completion”.

For the publishing industry, these concerns are not new but the recent Macmillan US issue does act as a salutary reminder to authors of the risks inherent in AI use and raises valid questions about how AI risk can be mitigated and managed by publishers, agents and authors alike.

Relevant principles of copyright law

In UK copyright law, the starting point is Section 1(a) of the Copyright, Designs and Patents Act 1988 (“the Act”) which refers to “original literary, dramatic, musical or artistic works”. In practical terms, the standard for “originality” is that the work in question is the author’s own intellectual creation. This definition can become problematic in the wake of AI, with generative AI potentially being used to draft, edit or enhance manuscripts and calling the “originality” of the author’s work into question.

UK copyright law goes some way to addressing at least the AI authorship issue via Section 9(3) of the Act, which provides that in the case of literary, dramatic, musical or artistic works which are computer generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken. As the law currently stands, the person “by whom the arrangements necessary for the creation of the works are undertaken” is usually the operator or the person directing the operation of the machine/software. In the case of an author using generative AI to draft, edit, or enhance a manuscript, the author would, then, be the person directing the operation of the machine – that is to say, the person giving the AI tool the prompts in order to do the drafting, editing or enhancing.

That does however leave open the question of originality and in turn the question of whether copyright subsists in work that has been generated, or partially generated, using AI – in particular, whether and in what circumstances the use of AI means that the output fails to meet the “originality” threshold. It is for this reason that publishers and agents are cautious about works that are under suspicion of having been drafted using AI – if the originality of the work is called into question, then so too is the copyright subsisting therein, meaning that the value of the work is compromised (the copyright subsisting in the works being the value) and the publisher is left with a manuscript it cannot commercially exploit.

AI policies

How, then, can this risk be managed? The obvious and most practical answer is for publishers to adopt robust AI policies, which make it clear amongst other things:

• to whom the policy applies (authors; illustrators; chapter contributors; etc.);
• the type of work to which it applies (text and illustrated work; fiction and non-fiction; academic and professional books);
• the types of use that are acceptable (e.g., for idea exploration; conducting preliminary research and brainstorming; for the copyediting of original works);
• the difference between using AI as an assistance tool versus a generative tool;
• the difference between public and private AI systems and what sort of use is acceptable within each system; and
• that the author must disclose the use of AI tools and assume responsibility for the originality and integrity of the content submitted to the publisher.

The required disclosure and assumption of responsibility would effectively act as a warranty, allowing the publisher to terminate the contract (and withdraw from the book deal) if the warranty is breached. This safeguards publishers, at least to some extent, from the risks associated with AI use in publishing insofar as they may be able to terminate contracts and clawback any advances. There are however practical issues that remain and which the recent Macmillan US matter highlights. In particular:

• How does an author establish the ways in which he or she has utilised AI?
• Who determines whether the use falls within the scope of the relevant AI policy?
• Where does an author draw the line between using AI as an “assistance” tool versus a “generative” tool?
• And how does the author establish originality in circumstances where he has availed himself of an AI tool?

There are no easy answers here but as a practical starting point, authors should keep records of their “conversations” with generative AI tools as well as records of the various iterations of their work. This information should, where appropriate, be shared with agents and with publishers in order to demonstrate that the output is the author’s own creative expression and in turn that the originality threshold is met. It is of course only where the originality threshold is met (and is beyond doubt) that a publisher will be satisfied that copyright subsists in the work and, in turn, that the work can be commercially exploited.

 Other risks

In the case of the Macmillan US deal, the author has strenuously denied the allegations of AI use and seemingly is seeking to pull together evidence of originality. The accusations are however in the public realm, with the author’s agents making a statement and the story widely reported in publishing news. This brings into play another risk flowing from AI use and that is the potential defamation argument that might be advanced by an author facing allegations of such use.

In Scots law, defamation is defined as the publishing of a statement which has caused, or is likely to cause, serious harm to the reputation of another (that is, if it tends to lower the person’s reputation in the estimation of ordinary persons). Similarly, in England and Wales, libel relates to defamatory statements that have caused, or are likely to cause, serious harm to reputation. It is fair to say that for authors, allegations of AI use may cause reputational harm sufficient to meet the “serious harm” test.

As such, publishers and agents should exercise caution in publicising concerns about AI use, lest they find themselves facing an action of defamation or libel which they then require to defend.

Risk mitigation

The recent Macmillan US incident is not the first of such incidents and it will not be the last. What the recent incident does do is offer publishers, agents and authors an opportunity to reflect on the use of AI within publishing generally and how that risk can be managed.

For publishers, that may look like putting in place AI policies which make it clear what is their position in relation to the use of AI and include a provision in terms of which the author effectively warrants that the work meets the originality threshold. If it then transpires that the policy has been breached, publishers can terminate the contract and look to recover any advance paid. Publishers should however exercise caution when suspicions of AI use arise and should seek to deal with these privately rather than via any sort of public statement, given the potential defamation/libel risk that attaches to a public statement.

Similarly, authors should exercise caution when utilising AI, ensuring adherence to any relevant policy and carefully documenting both the use of AI (via screenshots if appropriate) and by saving copies of all iterations of their work. On the face of it, this should enable an author to demonstrate originality and to answer any questions that arise about AI use.

Difficulties are still likely to arise when determining AI assistance versus AI creation, and where an author and an AI tool have both materially contributed to any given work, the question of originality does become harder to answer. Authors should be prepared to answer questions on AI use and, as above, to clearly demonstrate such use so that agents and publishers can authenticate how a manuscript has evolved from origin to completion.

Any mitigation strategy that is adopted will however require to evolve in line with AI technologies, which are ever-changing. In the meantime, clear policies, clear contractual drafting and proper record-keeping offer the best practical protection for publishers, agents and authors navigating the use of AI within publishing.

If you are an author, agent or publisher based in Scotland and looking for advice in relation to publishing contracts, AI policies or copyright law, please contact Lauren McFarlane at lmf@bto.co.uk or 0131 222 2944.