How is one AI catastrophe event defined for insurance and reinsurance?
sebbi.pro publishes a sealed, versioned AI Event Definition (SBP-AIE v1.0): an AI Loss Occurrence is all individual losses arising from the same originating AI cause and affecting AI systems built on the same model family, within a chosen period of consecutive hours (72 by default) starting at or after the first loss. The model family of each loss is read from sealed, Bitcoin-timestamped model lineage records, so it cannot be assigned after the event. An hours-clause engine groups a list of losses into occurrences and applies retention and limit per occurrence, and seals the result.
How it works
- Read the definition at sebbi.pro/.well-known/ai-event-definition.json.
- List the losses: insured, date and time, AI model family, amount and cause.
- The engine groups them by model family and cause into occurrences within the hours window.
- It applies the retention and limit to each occurrence and seals the report (AG-).
Catastrophe wordings define a windstorm or an earthquake so that losses can be aggregated for event limits and reinsurance recoveries. AI had no such definition. Wordings can cite SBP-AIE v1.0 and the clause SBP-AI-02 by their sealed fingerprints.