As generative AI enters the mainstream, crowdfunding platform Kickstarter has struggled to formulate a policy that satisfies parties on all sides of the debate.

Most of the generative AI tools used to create art and text today, including Stable Diffusion and ChatGPT, were trained on publicly available images and text from the web. But in many cases, the artists, photographers and writers whose content was scraped for training haven’t been given credit, compensation or a chance to opt out.

The groups behind these AI tools argue that they’re protected by fair use doctrine — at least in the U.S. But content creators don’t necessarily agree, particularly where AI-generated content — or the AI tools themselves — are being monetized.In an effort to bring clarity, Kickstarter today announced that projects on its platform using AI tools to generate images, text or other outputs (e.g. music, speech or audio) will be required to disclose “relevant details” on their project pages going forward. These details must include information about how the project owner plans to use the AI content in their work as well as which components of their project will be wholly original and which elements will be created using AI tools.In addition, Kickstarter is mandating that new projects involving the development of AI tech, tools and software detail info about the sources of training data the project owner intends to use. The project owner will have to indicate how sources handle processes around consent and credit, Kickstarter says, and implement their own “safeguards” like opt-out or opt-in mechanisms for content creators.

An increasing number of AI vendors offer opt-out mechanisms, but Kickstarter’s training data disclosure rule could prove to be contentious, despite efforts by the European Union and others to codify such practices into law. OpenAI, among others, has declined to reveal the exact source of its more recent systems’ training data for competitive — and possibly legal liability — reasons.