AI safety + compliance for insurance
The eval + guardrail + red-team + audit platform for insurers facing the wave of state AI regulation (NAIC, NYDFS, Colorado DOI, Connecticut). Catch unfair-discrimination patterns before they reach a state market-conduct exam, prove adverse-action explanations are grounded, ship the audit trail your regulator expects.
What ships today
Honest posture, not roadmap promises
Every checked item is in production today. In-progress items are flagged explicitly — no overclaiming, no vapor.
Built for buyer reality
Insurance AI use cases we ship for
Underwriting decision-support
AI scores submissions for risk + flags ones needing human review. Decision pipeline must satisfy state bias testing (e.g. Colo DOI 10-1-1), every accepted/declined account must have a citable rationale.
EvalGuard features
- Bias scorer pack: race, gender, age, ZIP-proxy testing per NAIC + Colorado DOI
- Decision-explanation faithfulness scorer (every rationale citable to underwriter rules)
- Audit log: every score linked to model version + ruleset version for market-conduct exam
- Per-state rate limits + multi-model routing for cost-sensitive direct-write personal lines
Claims-triage automation
AI assigns initial claim severity + routes to the right adjuster. Cannot deny a claim, cannot leak claimant PII, cannot give legal-style coverage opinions.
EvalGuard features
- PII firewall: claimant SSN, policy number, accident-location auto-redacted from logs
- Topic firewall: blocks responses crossing into coverage-determination territory
- Tamper-evident audit log: every triage decision survives the litigation hold
- Cross-tenant RLS: multi-MGA / multi-state isolation enforced at every query
Customer-service chatbot
Policyholder chatbot answers billing + endorsement questions. Cannot bind coverage, cannot make rate quotes that violate state-filed rates, cannot leak prior policyholders' data.
EvalGuard features
- Topic firewall: blocks queries asking for binding coverage quotes
- Hallucination scorer: faithfulness against the loaded state-filed rate corpus
- PII firewall: policy number, claim ID, banking info auto-redacted from logs
- Output toxicity scorer: catches dismissive phrasing on claim-denial conversations
Regulatory + ESG reporting
AI drafts NAIC-quarterly filings + sustainability reports from underlying claim + investment data. Every number must reconcile, every narrative must be diffable for state regulator review.
EvalGuard features
- Citation-format scorer: % of sentences with traceable source data points
- Numeric-faithfulness scorer: every number reproducible from underlying data
- Version-controlled prompts + outputs (prompt IDE) for filing-amendment defense
- EU AI Act high-risk classification for cross-border carriers
Wire it in 60 seconds
Wrap your OpenAI client. Scope to a state + line of business.
Per-state project IDs isolate audit + bias scorer thresholds + retention. Aequitas-grade bias deltas + adverse-action faithfulness configured in the control plane.
import OpenAI from "openai";
import { wrapOpenAI } from "@evalguard/openai";
const openai = wrapOpenAI(new OpenAI(), {
apiKey: process.env.EVALGUARD_API_KEY!,
projectId: `adverse-action_${state}_${lob}`,
metadata: {
vertical: "insurance",
state, lob,
regulators: ["NAIC", "Colo DOI 10-1-1"],
},
blockOnViolation: true,
evalOnResponse: { failOnScore: 0.85 }, // bias delta + faithfulness gate
});
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: claimDenialDraftPrompt }],
});wrapOpenAI for wrapAnthropic.Stack
Six surfaces, one platform
Eval, firewall, red-team, audit, BYOK, dashboard — every surface ships out of the box. No bolt-on vendors, no procurement cycle per capability.
Ship AI that survives the next market-conduct exam.
Free trial includes bias scorers, adverse-action faithfulness, NAIC control mapping, and the full audit-trail engine. State-specific evidence bundles on request.
Apache-2.0 source · SOC 2 evidence engine live · full trust center