The Failure Library grows from publicly documented AI incidents, contributed and reviewed the way an open reference work is built. These guidelines explain what belongs, how to write a good entry, and what happens after you submit. They take about two minutes to read — please skim them before contributing a case.
The Library documents governance failures involving AI and automation — moments where a system was granted authority beyond the evidence, context, or judgment available to support it, or where automation displaced human judgment in a way that caused harm. The lesson is the point, not the blame. If an incident helps leaders recognize a recurring failure pattern before it repeats, it fits.
A case is a good fit when it is:
Aim for the tone of a neutral case note: factual, concise, and focused on the governance breakdown. You do not need to be an expert or map the framework perfectly — the editors do the framework mapping and verification. A few sentences of accurate description with a good source is more useful than a long, unsourced account.
Nothing you submit is published automatically. Every submission is reviewed and verified before it can appear.
Your case enters an editorial review queue. It is not visible on the public catalogue.
Editors check the account against its sources and assess whether it fits the Library's scope.
Suitable cases are developed into a draft entry, mapped to the WEKID layers and Decision-Matrix outcome. Editing for accuracy and framing is normal.
If it holds up, it is published and credited to you. If it does not fit, it is declined. We cannot promise every submission is published.
When you submit a case, you agree to a short set of contributor terms. In plain language:
You confirm the contributor terms on the submission form.