The eval + guardrail + red-team + audit platform for AI workloads in banks, brokers, asset managers, and fintech. Block account-data leakage at the gateway, prove every customer-facing response was screened, generate the audit trail regulators expect.
What ships today
Every checked item is in production today. In-progress items are flagged explicitly — no overclaiming, no vapor.
Built for buyer reality
Retail-banking chatbot answers account questions + recommends products. Cannot give investment advice that crosses into licensed territory, cannot leak prior customers' balances, cannot output fabricated rates.
AI extracts beneficial-owner info from corporate-formation documents and flags suspicious patterns. False negatives expose the bank to BSA fines; PII handling must satisfy GLBA.
AI drafts the human-readable reason for a fraud-decline event so customer-service reps can communicate. Must not invent transaction details, must not leak the fraud-detection model's heuristics.
AI drafts CCAR / DFAST / 10-K narrative sections from underlying data. Every claim must be source-citable, every number must be reproducible, every draft must be diffable for compliance review.
Wire it in 60 seconds
PAN/SSN firewall patterns + ECOA-faithfulness scorer thresholds + FFIEC 7-year retention live in the EvalGuard control plane. Your code only wraps the client.
import OpenAI from "openai";
import { wrapOpenAI, EvalGuardOutputViolationError } from "evalguardai-openai";
const openai = wrapOpenAI(new OpenAI(), {
apiKey: process.env.EVALGUARD_API_KEY!,
projectId: "adverse-action-drafts",
metadata: { vertical: "finserv", regulators: ["FFIEC", "SOX", "GLBA"] },
blockOnViolation: true, // PAN/SSN never reach the model
evalOnResponse: { failOnScore: 0.8 }, // ECOA-faithfulness gate
onOutputViolation: (s) => routeToHumanReview(s),
});
try {
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: adverseActionPrompt }],
});
} catch (err) {
if (err instanceof EvalGuardOutputViolationError) {
// 7-year retention audit row already written for the FFIEC examiner.
}
}wrapOpenAI for wrapAnthropic.Stack
Eval, firewall, red-team, audit, BYOK, dashboard — every surface ships out of the box. No bolt-on vendors, no procurement cycle per capability.
Free trial includes PCI-DSS firewall, faithfulness scorers, evidence-bundle export, and the full audit-trail engine. SOC 2 Type II report on request once attestation completes.
Apache-2.0 source · SOC 2 Type II in progress · full trust center