AI safety + compliance for legal work
The eval + guardrail + red-team + audit platform for AI workloads in law firms, in-house legal, e-discovery, and legal tech. Block privileged-content leakage at the gateway, prove every research output is citation-grounded, satisfy ABA Model Rule 1.6 on confidentiality.
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
Legal & compliance AI use cases we ship for
Legal-research assistant
Associate asks the AI to summarize precedent on a specific cause of action. Every cited case must be real (not hallucinated), every quote must be verbatim, every Shepardization must be accurate.
EvalGuard features
- Citation-format scorer: blocks responses that name cases not in the verified corpus
- Hallucination guard: faithfulness check against the loaded research database
- Output guardrail: block mode rejects fabricated quotes or non-existent statutes
- Audit log: every research session tied to attorney + matter ID for billing + WP defense
Contract review + redlining
AI reviews an inbound contract against client-policy standards + flags deviations. Must not leak client policy positions to third-party LLMs, must explain every flag in citable terms.
EvalGuard features
- PII firewall: client name, party names, deal terms redacted from training-eligible logs
- BYOK keys: client-specific encryption isolates contracts per matter
- Output faithfulness scorer: every flag reasoned against policy-clause source
- Cross-tenant RLS: matter-isolation enforced at every DB query
E-discovery review
AI categorizes thousands of documents as responsive / non-responsive / privileged. False positives expose privileged material; false negatives expose the firm to sanctions.
EvalGuard features
- Privilege-detection scorer: trained on common attorney-client + work-product patterns
- Audit log: every classification linked to reviewer + cite-back to source doc snippet
- Output guardrail: block mode auto-quarantines high-confidence-privileged docs for human review
- Cost ledger: per-matter USD attribution for e-discovery cost recovery
Compliance-policy chatbot
Employees ask the AI about company policy on harassment / FCPA / export controls. Must never give legal advice, must always cite the policy section, must escalate the right way.
EvalGuard features
- Topic firewall: blocks queries that cross into the practice of law (UPL defense)
- Source-citation scorer: every answer cited to policy section + version
- Output toxicity scorer: catches dismissive or judgmental phrasing in harassment-related queries
- Audit log: privacy-preserving log of policy queries for compliance training analytics
Wire it in 60 seconds
Wrap your OpenAI client. Scope to a single matter.
Per-matter project IDs isolate audit + cost + scorer config. Privilege firewall patterns + citation-fidelity scorers + ABA-1.6 retention live in the EvalGuard control plane.
import OpenAI from "openai";
import { wrapOpenAI } from "@evalguard/openai";
// One wrapped client per matter — per-matter audit isolation,
// cost rollups, scorer thresholds.
const openai = wrapOpenAI(new OpenAI(), {
apiKey: process.env.EVALGUARD_API_KEY!,
projectId: `matter_${matterId}`,
metadata: { vertical: "legal", aba_rule: "1.6", privileged: true },
blockOnViolation: true, // privileged content gate
evalOnResponse: { failOnScore: 0.85 }, // citation-fidelity floor
});
await openai.chat.completions.create({
model: "gpt-4o",
messages: [{ role: "user", content: discoveryReviewPrompt }],
});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 your bar examiner could read in court.
Free trial includes the privilege firewall, citation scorers, and matter-isolated BYOK. Talk to us before sending the first piece of client work through any LLM.
Apache-2.0 source · SOC 2 evidence engine live · full trust center