The accumulation problem
A shared component can fail across several insureds. AI dependencies include models and versions, tools, identity providers, retrieval sources, hosting, deployment channels, and common policies. A shared update can alter behaviour; it does not inevitably produce simultaneous losses. STABLE
Build a dependency map
Section titled “Build a dependency map”Identify the shared component, which deployed versions use it, what business activities depend on it, what authority the systems have, the populations exposed, and the controls that could interrupt a loss. Then examine correlated effects and plausible severity. STABLE
An alert seen across deployments is a signal to investigate, not a measured portfolio loss. Declared model identifiers may need stronger provenance before being treated as verified facts. Privacy, consent, data quality, missingness and permissions also constrain aggregation. STABLE
The original brochure proposed a portfolio demonstration. It is not evidence that a production accumulation facility or reinsurer-accepted model exists. GROUNDED · Brochure
Continue
Section titled “Continue”→ Inference boundary · Accumulation and product design exercise
Drill this page →23 bank questions stand behind what you just read. Check it while it’s warm.