AI safety + compliance for government
The eval + guardrail + red-team + audit platform built to satisfy public-sector AI mandates. Map every model to NIST AI RMF + EU AI Act controls, generate the evidence bundle agency-CAIO offices require, run continuous red-team against the surface attackers actually exploit.
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
Government AI use cases we ship for
Citizen-services chatbot
Agency website chatbot answers benefit-eligibility questions across SSI, SNAP, VA, Medicare. Cannot promise benefits that don't exist, cannot leak PII across sessions, cannot give legally-binding determinations.
EvalGuard features
- Topic firewall: blocks queries asking for legally-binding eligibility determinations
- Hallucination scorer: faithfulness check against the loaded benefit-program corpus
- PII firewall: redacts SSN, DOB, household income from logs by default
- Audit log: every conversation tagged with agency + program for OIG audit recoverability
FOIA response triage
AI categorizes FOIA requests by exemption applicability + drafts initial response. Must never disclose exempted material, must cite every exemption to the FOIA statute, must satisfy 20-day response clock.
EvalGuard features
- Exemption-detection scorer: trained on FOIA b(1)–b(9) patterns
- PII firewall: cross-check on every release for unintended PII inclusion
- Output guardrail: block mode auto-quarantines high-confidence-exempt material
- Audit log: every triage decision linked to reviewer + statute citation
Acquisition / contract review
AI reviews vendor responses to RFPs + flags evaluation criteria deviations. Procurement-sensitive material must stay inside the firewall, FOIA-discoverable material must be auditable.
EvalGuard features
- BYOK keys: agency-specific encryption isolates RFP responses per acquisition
- Procurement-integrity scorer: catches contractor-bias language patterns
- Tamper-evident audit log: integrity_hash chain for protest-defense
- Cost ledger: per-acquisition USD attribution for indirect-cost recoverability
Mission-system observability
Agency-internal AI assists analysts on classified/CUI-marked workloads. Every prompt must be CUI-handling-compliant, every model output must be classification-marked, every interaction must be auditable to a clearance.
EvalGuard features
- CUI-marking scorer: flags responses lacking the required classification marking
- Multi-frame voice guardrails for SCIF-deployable voice assistants
- Per-user audit trail keyed to clearance ID (no joint-account misuse)
- Self-hosted Apache-2.0 source — full deployment behind your firewall
Wire it in 60 seconds
Wrap your OpenAI client. Point it at your air-gapped EvalGuard.
Self-host inside the ATO boundary by pointing `baseUrl` at your private deployment. NIST RMF + EU AI Act + ISO 42001 evidence is generated from the audit stream.
import OpenAI from "openai";
import { wrapOpenAI } from "@evalguard/openai";
const openai = wrapOpenAI(new OpenAI(), {
apiKey: process.env.EVALGUARD_API_KEY!,
baseUrl: "https://evalguard.agency.gov/api/v1", // air-gap deploy
projectId: "benefit-determination",
metadata: { vertical: "government", risk_tier: "high", classification: "CUI" },
blockOnViolation: true,
evalOnResponse: { failOnScore: 0.8 }, // eligibility-faithfulness gate
});
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: benefitDeterminationPrompt }],
});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 public-sector AI that survives the IG audit.
Free trial includes the full NIST + EU AI Act control mapping + evidence-bundle export. Self-hosted deployment + FedRAMP attestation path on request.
Apache-2.0 source · SOC 2 evidence engine live · full trust center