Chaosium removes more than 40 community titles in crackdown on AI use

Chaosium has removed more than 40 titles from sale across its community content programmes, saying the publications contained AI-generated text. In an update published on 21st October, the company says its ambassadors will continue reviewing work they believe shows signs of AI use, and warned that affected creators’ future publications would also be removed.

The programmes named in the announcement are the Miskatonic Repository for Call of Cthulhu, the Jonstown Compendium for RuneQuest and Glorantha, Companions of Arthur for Pendragon, and Sagas of the North. These community programmes let creators publish material for Chaosium’s games through DriveThruRPG.

Chaosium says its review process combines staff experience with several AI-detection tools, highlighting a variety of clues they look for, including unusually uniform formatting, a relentlessly breathless present-tense style, errors in the application of game rules or setting lore, and scenarios that are lengthy but difficult to run. It also singles out very rapid publishing schedules as a reason to investigate. The company has not named the creators or titles involved, but has said that none of the affected authors belonged to its Facebook community writers’ groups.

We use a mix of our own instincts from decades of writing scenarios and running games, backed up the use of multiple AI detection tools of we smell a digital rat. Volume of publication is another obvious indicator that we should take a look. We know many of you and you often share what you are working on, how it’s going, previews of the cover, when it comes out: you all know that the process takes weeks, often months, sometimes years. But when a brand new author appears out of nowhere and publishes one or more new scenarios daily, our suspicions are aroused, and soon confirmed on inspection.

The announcement’s advice to creators is quite simply: “Just don’t use AI.”, telling contributors not to use AI to generate ideas, write or revise scenario text, summarise or paraphrase material, check rules, create statistics, or produce artwork, layout, blurbs or covers, cautioning that tools such as Microsoft Copilot and Google Gemini, as well as AI-powered autocorrect suggestions, may alter a writer’s voice.

This phenomenon has only risen in recent times, and all of you here who make honest effort have nothing to worry about. If you are concerned, here's what not to do when preparing Chaosium Community Content publications:

— Don't use AI to generate ideas for a scenario. Those festering digital bones will remain visible though to the final publication.

— Don't use AI to write the actual scenario text, obviously.

— Don't use AI to summarise, paraphrase, check rules, or create stats. It will get them wrong, or it will regurgitate them, or it will express them in a weird way.

— Don't use AI to revise or check your human text. It will rewrite it into AI-speak. Be wary of Microsoft Copilot and Google Gemini here, and anything that makes autocorrect suggestions for you, as it replaces your voice with AI. We strongly recommend disabling such tools from your workflow.

— Don't use AI to do your artwork.

— Don't use AI to do your layout.

— Don't use AI to write your blurbs.

— Don't use AI to create your covers.

— Just don't use AI.

This goes further than the policy Chaosium set for AI-generated artwork in 2023, which barred AI art from new community titles. DriveThruRPG’s published guidelines also state that commercial marketplace content primarily written by AI language generators would not be accepted from July 2023. Chaosium’s new statement takes a more sweeping position on AI’s role across the whole creation process.

The announcement leaves questions about how individual cases are assessed and what recourse creators have if their work is challenged. Chaosium lists stylistic and production clues alongside the use of detection software, but does not describe an appeals process in the post. Its examples, such as polished grammar, em dashes, tidy layouts and frequent tables, are not, by themselves, guaranteed 100% proof that a work was generated by AI. The company’s stated position is that suspicious work is inspected, and that publications displaying the characteristics it associates with AI writing will be removed.