The translation layer for AI insurance
RiskBorne turns a carrier's intake into something an applicant can act on — and turns the applicant's evidence into what the carrier needs to quote.
covered · conditional · excluded · insufficient evidence
Who it's for
Your carrier asked for ground truth, an error rate and a dataset nobody on your team knows how to produce. We translate the question, help you define what error means for your model, and collect the evidence with you.
You trigger it, the applicant's technical contact does the technical part, and you get back a package you can forward to market.
Your template, your format. Packages arrive shaped for your quote engine, with the evidence behind every answer.
The forcing event
As those endorsements land, AI liability stops sitting with the policy and starts sitting on the enterprise balance sheet. RiskBorne is the bridge back to coverage: evidence of what your agents do, measured against what carriers now require.
How it works
Upload the form your carrier sent. We turn it into plain-language questions your technical contact can actually answer.
Say what a wrong answer means for each model, then connect the agent or upload a sample so the error rate is measured, not asserted.
The carrier receives their own format, ready to rate — and it is kept current as models and evidence change.
The coverage gap map
The gap map is where a risk committee stops arguing and starts working. Agents run across the top, carrier requirements down the side, and each cell carries a verdict you can trace back to the clause that produced it.
Illustrative example · coverage gap map
index 66 · grade D
| Requirement | Patient Intake | Claims Triage | Provider Copilot |
|---|---|---|---|
| A written AI usage policy is in force across the organizationCarrier A · Condition 1 | Coveredsatisfied | Coveredsatisfied | Coveredsatisfied |
| Vendor safety guardrails must remain unmodifiedCarrier A · Condition 2 | Coveredsatisfied | ExcludedCondition 2 | Coveredsatisfied |
| Privacy notices reviewed by counsel where protected data is processedCarrier A · Condition 3 | Coveredsatisfied | InsufficientCondition 3 | InsufficientCondition 3 |
| Drift detection with a scheduled model refreshCarrier B · Monitoring duty | Coveredsatisfied | ConditionalMonitoring duty | ConditionalMonitoring duty |
| Role of the model in the decision path is disclosed at underwritingCarrier B · Disclosure duty | Coveredsatisfied | ConditionalDisclosure duty | Coveredsatisfied |
| Human review stands between output and irreversible actionCarrier C · Scheduled condition | Coveredsatisfied | ExcludedScheduled condition | ConditionalScheduled condition |
Why insurance-led, not just governance
Most AI governance tools sell policy documents to compliance teams. RiskBorne starts from the other end: the requirements carriers actually underwrite against. Because every question we ask traces back to a requirement someone is willing to insure, the output maps to real coverage rather than to a framework — and that is the difference between a binder you can rely on and a file that reads well.
Carrier coverage
The requirement corpus is transcribed from published wordings and application questions from the carriers and MGAs writing affirmative AI cover. Each requirement is stored with its clause reference, the evidence it expects, and the version it came from, so a verdict can always be traced back to a source. RiskBorne is independent: naming a carrier here does not imply any partnership, appointment or endorsement, and no carrier pays for how a verdict lands.
Armilla
Affirmative AI liabilityTestudo
Affirmative AI liabilityaiSure
AI performance coverHSB
AI endorsement coverEvidence
RiskBorne accepts OpenTelemetry, the open standard the industry has converged on — so evidence flows from the tools your applicants already use, without custom integration work on either side. Datadog, Langfuse, Arize, Braintrust, Weights & Biases, Honeycomb and whatever ships next all export OTLP. For anything that does not, there is a signed webhook and a polling endpoint.
We read monitoring metadata only — whether monitoring exists, when it last ran, error counts, and model or version changes. Never prompts, completions or customer data. We store aggregates, not spans.
OpenTelemetry guide →Webhook and heartbeat →
Webhook and heartbeat available today · OpenTelemetry coming soon
Questions
Whether an insurer will cover the losses your AI agents can cause. Carriers writing affirmative AI cover publish conditions — guardrails left intact, human review in the decision path, drift monitoring, disclosure to affected people. Insurability is the distance between how your agents actually run and what those conditions require.
Request access and get a per-carrier insurability verdict with the exact gaps blocking coverage.
