Verticals / 2.4

Insurance & insurable-by-design.

Reproducibility equals insurability. The forensic chain of custody is what allows an insurer to underwrite a governance layer at all.

01Forensic chain of custody as the insurance foundation

An insurer underwriting an institutional governance layer is asked to price a loss model. A loss model requires a tractable distribution over outcomes — given an action, what is the distribution of consequences, and how does the layer's behaviour shape that distribution? When the layer is probabilistic, the loss model is itself probabilistic-on-probabilistic: the insurer is pricing a guess about a guess.

H.A.R.I. removes the inner guess. Every governed action carries a forensic chain entry: the inputs, the policy version, the canonical output, the named authorizing human if applicable. The loss model becomes tractable because the layer's behaviour, conditional on inputs, is exact rather than statistical.

02Reproducibility equals insurability

The institutional headline is short: reproducibility equals insurability. The shorthand carries weight because it inverts the usual posture for AI-adjacent systems, where insurance is treated as something that gets bolted on once the technology is "good enough." Here, the architectural property that makes the system reviewable is the same property that makes it underwritable.

For underwriters

The chain provides the ground truth. The policy version provides the operative rules. The replay procedure (read inputs → apply policy → compare output) is the same procedure an underwriter would write into a claim-investigation playbook. The audit surface and the underwriting surface are the same surface.

03Audit-ready architecture for insurers

Insurers do not need vendor-specific tooling to interrogate the platform. The forensic chain is queryable, the records are signed with audit-grade ECDSA P-256, and the OpenTimestamps proof — current ID SOV-2026-02-26 — anchors the chain in a public, third-party-independent timeline. The data model the underwriter receives is the same data model the regulator receives.

04Differentiator vs. probabilistic AI systems

For a probabilistic AI system serving a comparable use case, the insurer must price the residual risk that the system's tail behaviour is unknown. There is no replay procedure that produces the same answer twice. The insurer can sample, run synthetic stress tests, and price conservatively — but the structural property that makes pricing precise is absent.

For H.A.R.I., the structural property is present. Tail behaviour is specified by policy. The insurer prices what the policy says, conditional on the chain delivering the inputs and the layer delivering the output. The two architectures are not substitutes; the insurance relationship is the most direct evidence of the architectural difference.

Status

Architecture-validated for insurance-aligned governance. Pilot-ready when scoped with an insurer or insurance partner whose mandate covers the deployment.