Sooner or later — usually the week after someone notices a suspiciously fluent press release — every arts organisation arrives at the same conversation: we should probably have a position on AI. Most then do one of two bad things. They produce nothing, and let practice drift into whatever each staff member privately does. Or they produce eleven pages of borrowed corporate language that nobody reads and artists actively distrust. There is a third option. It's one page long, and writing it might be the most useful meeting you have this year.
Why artists hate most AI policies
Because most AI policies are about permission, and artists' fears are about displacement. A document that spends eleven pages on approved tools and password hygiene, and no sentences on whether the organisation will generate images instead of paying illustrators, has answered the wrong question with impressive thoroughness. Your artists don't need to know your stance on browser extensions. They need to know whether you're on their side — and vagueness reads as a no.
The one-pager
Six headings. Plain English throughout — if a sentence wouldn't survive being read aloud at a company meeting, cut it.
- Why we use AI at all. Two sentences, honestly stated. Something like: "We use AI to reduce administrative work so more of our time and money goes to artists and audiences. We do not use it to replace creative work we would otherwise commission."
- What we use it for. The green list, concretely: minutes, summaries, first drafts of admin documents, reformatting, access materials, spreadsheet work. Real examples beat categories — borrow ours if useful.
- What we never use it for. The red list, and the heart of the document: generating artwork, images or music in place of commissioning artists; cloning or simulating any person's voice or likeness; making decisions about people (recruitment, casting, participant selection); writing anything that carries the organisation's feeling — condolences, crisis responses, artistic statements. Draw the lines that match your values; just draw them in ink.
- What we tell people. The transparency clause: machine-generated content is labelled, colleagues aren't misled about authorship, and if a funder or artist asks how something was made, they get a straight answer. (The test we use for every grey area: would we be happy to say out loud how this was made?)
- What never goes into the machines. Personal data about identifiable people — donors, staff, participants — plus anything confidential or under embargo. One organisational account with checked settings, not a scatter of personal logins. (The reasoning is here.)
- Who's accountable. A named human checks and owns everything that leaves the building, machine-assisted or not. "The AI got it wrong" is not a sentence this organisation will ever say to a funder.
Add a review date every six months — this field moves fast, and a policy with a sell-by date printed on it earns more trust than one pretending to be eternal.
How to write it (this is the part that matters)
Don't draft it in a leadership huddle and hand it down. Put the six headings on a wall, invite the whole team and some of the artists you work with, and fill in the lists together. The conversation will surface exactly the anxieties and grey areas the policy exists to settle, the red list will be stronger for containing the things your artists actually fear, and — the quiet magic — people defend documents they helped write. An hour's workshop, an afternoon's tidy-up, done.
Then publish it. On the website, not the intranet. An organisation confident enough to state its lines in public is making a promise it can be held to — which is precisely what makes the promise worth something. In a sector where trust in institutions is the whole game, one honest page beats eleven careful ones every time.
The image above was AI-generated — as ever, we'll always tell you.