Society of Authors withdraws from its own ethical AI charter after member backlash

What the Society of Authors decided

The Society of Authors (SoA), the UK trade union for authors, illustrators and translators, has withdrawn from the Charter of Ethical AI Use in the Creative Industries that it launched barely a week earlier. Members were emailed on the morning of 21 October 2026 to thank them for their feedback and to tell them the SoA would be withdrawing, The Bookseller reported.

Chief executive Anna Ganley told the trade paper: "I have spoken with all partners. We will pause, listen to members' views and take these back to our partners." The next steps will be discussed at the SoA's annual general meeting on 19 November, and members are being invited to share their views in a survey.

The reversal is narrower than it may sound. The charter itself is still live, and according to The Bookseller no other organisation had withdrawn from it at the time of reporting. It was created with the Creators' Rights Alliance in the UK, the Authors Guild in the US, the Australian Society of Authors, the Writers' Union of Canada and the New Zealand Society of Authors. What has changed is that the organisation that led it has stepped back after an unusually sharp reaction from its own members.

What the charter asked publishers to do

The charter, published on 13 October according to The Bookseller, was aimed at organisations rather than individual writers: agents, publishers, producers, prize bodies and trade associations. Its stated purpose was to "define how generative AI services should be deployed to help protect and support human creators" while governments around the world were still not committing to AI regulation and oversight. It describes itself as a voluntary code, "a small but significant step" that will not solve the larger threats that unregulated generative AI poses.

Signatories commit to four headline principles: promoting responsible and ethical use of generative AI services, following UNESCO's 10 core principles of responsible AI and the CLEAR framework, making sure that "where AI is used, it is used for good", and working mindfully in relation to the impact on jobs, livelihoods and the environment. CLEAR stands for consent, licensing not scraping, ethical use, accountability, and rights and remuneration. It was adapted from Brave New World?, a report on creators and generative AI published by the SoA with the Independent Society of Musicians, Equity, the Association of Illustrators and the Association of Photographers.

The practical section goes further. Signatories promise to train their staff, risk-assess any proposed use case for its effect on IP, data protection and creators, write and publish an AI permitted use and disclosure policy, seek consent before using a creator's work to train a model or to create a new work from it, and be transparent when AI-generated content forms a material part of a published work. Where viable alternatives exist, they say they will favour AI services from developers who license their training data. The charter also asks creators to tell organisations when generative AI has materially shaped their own work. When we read the page on 21 October, its signatory list showed Alma Books and the Australian Society of Authors.

Why members objected

Criticism came on two fronts, The Bookseller reported: the lack of consultation with members before the charter was published, and what the document implied about the SoA's position on generative AI. Writers who have spent three years watching their books turn up in training datasets read a pledge to use AI "for good" as a concession that ethical use is possible at all.

Romance novelist KJ Charles set out that case in a blog post that reproduced her letter to Ganley. She said the authors' bodies behind the charter had not consulted their memberships and had "signed it in our names". She agreed with the SoA that LLMs were trained on copyright work without permission or payment, but argued that the charter "offers publishers a way to virtue-wash their GenAI use" without any checks, measurable goals or consequences. She called the approach "jawdroppingly naive" and said her resignation from the society stood.

Her alternative was a harder line: a union that names publishers using generative AI to replace creatives, pushes for a class action against what she called the book thieves, and insists that authors be made whole before there is any talk of publishers using LLMs ethically. She also wrote that the response on Bluesky had been overwhelmingly hostile. That argument is not unique to publishing; in another profession, mathematicians have been signing a pledge not to work for AI companies at all, as we reported when the Association for Human Mathematics passed 1,000 members.

The CEO's defence

Ganley first answered the criticism in a CEO blog dated 15 October (The Bookseller dates the statement 16 October). "I am sorry that we clearly failed to communicate its purpose," she wrote, adding that the SoA's position on generative AI "remains unchanged". The society, she said, regards generative AI as harmful to the creative industries, to writers' livelihoods and to the environment, and does not encourage its use.

"We are clear that there can never be any truly ethical use of generative AI because of the way LLMs have been trained on copyright works without permission or payment; the detrimental environmental impacts; and the harmful effect it has on disempowered communities," she wrote.

Her case for the charter was pragmatic. An increasing number of publishing organisations are already using generative AI, she wrote, and the SoA was "deeply concerned that people within the industry are using generative AI in the shadows, where it cannot be monitored". The charter was meant to extract "public, transparent commitments from all publishers about whether, how and where generative AI is being used", and to make sure that where organisations do see a case for it, such as data analysis and summarisation, it is used in a way that limits harm. Critics like Charles accepted the diagnosis but not the remedy: they argued that a pledge is only as strong as the willingness of its signatories to keep it.

What happens next

For now the SoA is collecting member feedback through a survey and will discuss the charter at the AGM on 19 November, which Ganley had already promised in her blog before the withdrawal. The charter text says it will be reviewed annually with input from signatories, and that organisations can remove themselves from it at any time. It does not say what happens if the body that wrote it leaves.

Ganley also used her comments to The Bookseller to restate the society's campaign priorities. "We are clear that authors' intellectual property rights must be protected and respected in this age of AI," she said. "Authors must have meaningful choice, consent and control over their work." She added that the industry is built on human writers, illustrators, translators, audiobook narrators and photographers, and that "we need to look after our creative workers, not the trillion-dollar US tech titans who exploit and extract from our UK books ecosystem".

She said the SoA is currently supporting the Australian Society of Authors by writing to Australia's attorney general and prime minister about the country's AI copyright proposals. According to Ganley, the Australian government is considering an opt-out scheme and a statutory licence or extended collective licence for AI training on material that is "unprotected" on the internet, and the SoA wants clarity on how that would affect international authors who publish work online.

Why a voluntary code is a hard sell

The episode shows how little room creative unions have between outright opposition and managed adoption. Courts are still working through the training-data question; in the US, a judge has ordered Meta to name everyone who approved downloading pirated books for Llama, while publishers and booksellers are making their own calls, from refusing suspicious bulk orders to using AI tools in their own work, which the SoA says is often happening out of sight. A voluntary charter tries to set terms for the second group without waiting for the first to finish.

The problem the SoA ran into is that any such code has to define acceptable use, and for many members the acceptable amount of generative AI built on unlicensed training data is zero. A union can lobby against a technology and set rules for how employers use it, but asking members to endorse both at once proved too much. Other creative unions face the same tension when they negotiate AI clauses with employers while lobbying against unlicensed training.

For publishers, the practical commitments in the charter, published AI policies, consent before using a creator's work and disclosure of material AI use, are things many authors have been asking for in contracts anyway. Whether they spread through a shared charter, through individual contract clauses or through regulation may now depend on what SoA members say before 19 November.

What to watch

Three things will show whether the charter survives the SoA's exit. The first is whether the Authors Guild, the Writers' Union of Canada, the Australian and New Zealand bodies and the Creators' Rights Alliance keep their names on it; as of 21 October none had withdrawn. The second is whether any publishers sign: on the day we checked, the public signatory list was short. The third is what SoA members decide at the AGM, and whether the society returns with a tougher document that includes the checks and consequences critics said were missing.

The charter is also a useful reference regardless of who signs it. Its CLEAR terms, consent, licensing, ethical use, accountability and remuneration, are the same principles creative groups are pushing lawmakers to adopt, and they map closely onto the disclosure and licensing demands in copyright cases against AI developers. The fight is less about those principles than about whether endorsing any framework for using today's models concedes too much.