Register of Anticipated AI Failure Modes

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This register collects twelve classes of AI failure that are documented today, in published research or in reproducible experiments, at a level of generality above any single system in the author's corpus. It exists because the author's own architecture documents — Verifiable Trust, the Charter of Procedural Standing, GVAA, BRAA, Physical Oracles, and the Local Multi-Scale Contour specification — each address a subset of these failure classes from within their own scope, and no one of them shows how the classes interact across system boundaries. The existence of this register reduces the probability that a failure class is simply forgotten. It does not reduce the probability that a failure class occurs, where the corresponding mechanism has not been tested. A well-organized register of failure modes, unaccompanied by a matching experimental program, raises the risk of institutional complacency. That risk is treated here as a failure mode of process, not only a property of the system under review.