Question 1

What is Verified Legal AI?

Accepted Answer

Verified Legal AI is Guardian Posse's multi-agent legal research system where every citation is independently re-derived — not assumed, not self-reported by the AI that generated it. Four or more independent AI models check every citation simultaneously against primary authority. Four consensus gates must all pass (Coefficient of Variation ≤ 1.0%, Modified Z-score ≤ 3.5, Fleiss' κ ≥ 0.75, Quorum). Every run is sealed into a SHA-256-chained cryptographic receipt. The system catches hallucinated citations, overruled law, and misquoted statutes before they reach a lawyer.

Question 2

How does multi-agent legal citation verification work?

Accepted Answer

Each legal citation is submitted simultaneously to 4+ independent AI models. Each model independently locates and verifies the primary authority — the actual statute, case, or regulation cited. The results are compared using four statistical gates: Coefficient of Variation (CV ≤ 1.0%), Modified Z-score for outlier detection (|Mᵢ| ≤ 3.5), Fleiss' κ for inter-model agreement (≥ 0.75), and Quorum (simple majority). If any gate fails, the citation surfaces as UNVERIFIED. Every run — pass or fail — is sealed with a cryptographic receipt any auditor can re-derive.

Question 3

What types of legal errors does Verified Legal AI catch?

Accepted Answer

Verified Legal AI catches: (1) Hallucinated citations — case names, docket numbers, or statutory sections that do not exist; (2) Overruled or reversed cases — citations to precedent that has since been overruled; (3) Misquoted text — citations that exist but where the quoted language differs from the actual holding; (4) Wrong jurisdiction — citations to law from a different state or circuit than the one applicable; (5) Superseded statutes — citations to statutory text that has since been amended or repealed. Verification reduces unsupported citations — it is not a guarantee of legal correctness.

Question 4

What is the chain of custody for legal research?

Accepted Answer

Every Verified Legal AI run produces a SHA-256-chained receipt that binds the query input hash, model IDs, verification verdicts, and timestamp. The chain is ordered, tamper-evident, and re-derivable — any auditor can verify that the cited output came from the stated verification run without trusting the operator or the AI. This chain of custody satisfies NIST AI RMF MEASURE 2.5 (independent trustworthiness evidence), EU AI Act Article 9 (risk management documentation), and SOC II audit trail requirements.

Question 5

How does Verified Legal AI relate to NIST AI RMF and EU AI Act?

Accepted Answer

NIST AI RMF MEASURE 2.5 requires measurement of AI system trustworthiness with independent evidence — Verified Legal AI's multi-agent consensus receipts satisfy this directly. EU AI Act Article 9 requires high-risk AI systems (which include AI used for legal interpretation) to implement risk management systems — the cryptographic receipt chain is the risk management record. Each run produces an Inference Audit Report (IAR) that maps to specific framework controls, making the compliance evidence auditable rather than self-reported.