1,847 documented AI hallucination decisions in courts worldwide

Your AI made a decision.
Can you prove it was defensible?

Independent AI verification for law firms. Citations checked against court databases, AI summaries checked against your private documents.

Attorneys who file AI-generated citations without independent verification face monetary sanctions, bar referrals, and permanent reputational damage. 1,847 documented decisions worldwide, 719 of them involving a practising lawyer. The rate is accelerating.

How VariLegal verifies Architecture and patents

Patent pending  ·  EU AI Pact Pillar II Signatory  ·  Built on open-source court data

The Regulatory Pressure

Judicial scrutiny of AI-assisted filings is accelerating.

1,847
Court decisions worldwide addressing AI-hallucinated content in filings
Charlotin Database, 6 August 2026
719
Of those cases where the party using AI was a practising lawyer, not a self-represented litigant
Charlotin Database, 6 August 2026
1,278
Cases in United States courts, of 1,847 identified across 40 jurisdictions worldwide
Charlotin Database, 6 August 2026
$15K
Monetary fine plus bar referral in a single July 2026 matter involving fabricated authority
CA Illinois, 28 July 2026
"Counsel's submission contained citations to cases that do not exist. The Court finds that this conduct falls below the standard of competence required under Rule 11."
— Federal district court order, sanctions proceeding
"The attorney represented that the cases cited were real and had been reviewed. They were not. They were generated by an artificial intelligence tool and have no existence in law or fact."
— Judicial sanctions order, 2024
"The submission of non-existent judicial opinions with false quotes and citations is a concerning development. The Court is duty-bound to respond."
— S.D.N.Y., cited in subsequent standing orders
Full sanctions reference →
Benchmark Results

Tested against a preregistered dataset.
Results locked.

100%
Fabrication detection, 45 of 45 AI-fabricated citations caught
Stanford RegLab preregistered dataset
0%
False leniency, no fabricated output passed as verified. Sole pass/fail gate for model upgrades.
Frozen Cloud Run revision
96.8%
Overruled citation recall, 217 of 225 overruled citations presented as good law, detected
Measured against open-source CourtListener data
Full Benchmark

Every number is reproducible.

Run against the Stanford RegLab preregistered dataset. Results locked at frozen Cloud Run revision. Not marketing claims, benchmark methodology.

Metric Result Sample Notes
Fabrication Detection 100% 45 / 45 cases Every AI-fabricated citation flagged. No fabricated output passed as verified.
Overruled Citation Recall 96.8% 214 / 221 cases Overruled citations presented as good law. Validated against the complete set of 225 overruled Supreme Court citations in the Stanford RegLab preregistered dataset, using only open-source CourtListener data.
False Leniency Rate 0% Full test set The sole pass/fail gate for any model upgrade. No verified output contained a material error.
Document Fabrication (v2) Detected Synthetic SPA test £4.5M purchase price fabrication, wrong Long Stop Date, wrong escrow period, and additional material errors caught in one pass.

the complete set of 225 overruled Supreme Court citations in the Stanford RegLab preregistered dataset. Frozen Cloud Run revision 00113. Built entirely on open-source CourtListener data. No commercial citator dependency. 0% false leniency is the sole pass/fail gate for model upgrades. 0% false positives on 48 verified citations. Three-stage waterfall enrichment: standard scan, deep cite scan (150 opinions), and corpus search recover treatment signals beyond the citation graph.

Rubicon — Document Verification Mode

Harvey summarized your contract.
Did it get it right?

Upload any private document. Paste the AI's output. VariLegal verifies every claim the AI made against the actual document text — and certifies what it got right, wrong, and omitted.

01
Upload your document

PDF or text. Contract, due diligence memo, policy, clinical document. Your document becomes the ground truth. It never leaves your control.

02
Paste the AI's output

The summary, extraction, or analysis Harvey, Legora, or CoCounsel produced from your document. VariLegal does not need access to those platforms — only their output.

03
Receive a verification certificate

Claim-by-claim analysis. Every finding traced to a specific passage in your document. SHA-256 hashed. Tamper-evident. Ready for the file.

No citator required. Document Mode operates entirely outside any citation database. Your document is the source of truth. No Westlaw. No CourtListener. No third-party data agreement.

What VariLegal caught on a real contract — today

A Harvey-style AI summary of a Master Services Agreement was verified against the source document. Six material errors found in one pass:

  • Payment due in 45 days — document says 30 days
  • Liability cap stated as 6 months — document says 12 months
  • Confidentiality survival stated as 3 years — document says 5 years
  • Termination for convenience — document says termination for cause
  • Liability cap presented as universal — three carve-outs omitted
  • 15% termination fee — fabricated, not in the document

Verdict: HALLUCINATION_DETECTED. Required action: REJECT.

Confidence score 0.25. The AI achieved 31% accuracy across 13 claims. Three financial terms were fabricated. Two numerical values understated by 40–50%. One legal standard mischaracterized. Certificate hash: 028d31f6f746a4a9...

The Certificate

What a partner actually receives.

Every VariLegal certificate is SHA-256 hashed at generation. Tied to the exact content verified, the agent verdicts, and the timestamp. Any subsequent alteration invalidates the hash.

VariLegal Certificate of Verification
DocumentVerity / Hartwell SPA
Verification ModeDocument Mode, Rubicon
Advocate VerdictVERIFIED
Adversary VerdictREJECTED
Arbitrator VerdictFLAGGED
System of RecordREJECTED
Critical Findings4 Material Errors
Confidence Score0.31 / 1.00
Timestamp2026-03-19T14:22:07Z
SHA-256 · a7f3c9d2e1b84506f2a3c7d9e0b14523f6a2d8c1e3b74509f2a1c8d7e0b34512

An immutable record of verification

When opposing counsel, a bar disciplinary panel, or a client asks whether the firm exercised documented oversight of AI-generated work product, the VariLegal certificate is the answer. Not a policy. Not a procedure manual. A record, with a hash.

Works alongside your existing workflow

VariLegal does not replace Harvey, Legora, or CoCounsel. It runs after them. Generate with the tool of your choice. Verify with VariLegal. Attach the certificate to the matter file.

EU AI Act compliance

Legal AI use is classified as high-risk under EU AI Act Article 22. VariLegal certificates provide the documented human oversight record required for compliance, structured, auditable, produced at the point of use.

What Was Built

A working verification service, not a concept.

VariLegal ran as a production service verifying citations against CourtListener's federal case database and returning a compliance certificate in under 90 seconds. The environment is currently held offline and can be restored for evaluation.

  • Citation existence verification against CourtListener
  • Holding accuracy validation and mischaracterization detection
  • Document Mode, verifying AI output against your private documents
  • SHA-256 hashed certificates per verification
  • Addresses NDTX, EDMI, and Fifth Circuit AI disclosure rules
  • Deterministic scoring, hard failures override LLM opinions

Two modes, one platform

Citation Mode verifies AI-generated citations against CourtListener's federal case database. Document Mode verifies AI summaries and extractions against your own uploaded documents, with no external citator dependency and no third-party data agreement.

Built on a commercial agreement with the Free Law Project. Benchmark methodology and results are published open access.

For Legal AI Platforms

Embed the verification layer. Deliver defensibility.

VariLegal operates as an independent middleware dependency between AI output and regulatory submission — not affiliated with, not replaceable by, the AI that generated the original output.

Enterprise law firms are demanding independent AI citation verification before filings. Platforms that can demonstrate independently verified, audit-logged citations win enterprise deals and reduce institutional risk.

The defensibility layer that legal AI platforms will need as judicial scrutiny intensifies.
Discuss API Integration
System Architecture — Independent Middleware Layer
AI Generation Layer
Harvey AI · Lexis AI · CoCounsel · GPT-4o · Any legal AI platform
VariLegal Verification Layer
Independent adversarial verification
Advocate · Adversary · Arbitrator
Regulated Outcome Layer
Court filing · Hiring decision · Insurance determination · Grid operations
VariLegal is not a feature of your AI platform. It is the independent verification dependency between AI-generated output and any regulated submission.

VariLegal does not bet on one model vendor. The Advocate, Adversary, and Arbitrator run a permanent benchmarking operation across model families. When a better model ships, it is evaluated, and the best combination is deployed. Every model improvement in the industry automatically becomes a VariLegal improvement.
For Law Firms

The partner who filed it is the one who gets sanctioned.

AI does not appear on the signature line. The attorney does. Independent adversarial verification is the documented record that stands between a partner and a sanctions motion.

Partner liability
The court does not sanction the AI. It sanctions the attorney who signed the filing.
Over 950 documented sanctions cases worldwide as of early 2026, at a rate of 2 to 3 new cases per day. In every case, the AI platform carried a disclaimer. In every case, the liability landed on the attorney. VariLegal produces a SHA-256 hashed compliance certificate documenting that every citation was independently verified before filing. That certificate is the due diligence record.
Client-facing defensibility
Your clients are already asking what your AI verification process looks like.
Enterprise clients with legal ops functions are auditing outside counsel AI usage. The question is no longer whether a firm uses AI. The question is what the firm does to verify it. A documented, independent verification process is the answer. Firms that cannot produce one are losing the conversation before the pitch begins.
Competitive differentiation
The firms that move first on documented verification own the standard.
AI adoption in legal practice is converging across the Am Law 100, Magic Circle, and Global Elite. The differentiator is no longer which AI platform a firm uses. It is which firm can demonstrate that its AI output was independently verified before it reached a client or a court. VariLegal is that demonstration.
Works with your existing stack
VariLegal verifies the output. It does not replace the tool that generated it.
Harvey, CoCounsel, Lexis AI, Westlaw AI, Copilot for Microsoft 365. VariLegal operates as an independent verification layer on top of whatever your attorneys are already using. No platform migration. No workflow disruption. Paste the AI output, upload the source document, receive a court-defensible verification certificate in under two minutes.

Verification is not a feature your AI vendor will build for you. It is structurally impossible for them to build it. An AI platform cannot independently verify its own output. Independence is the product. That is what VariLegal delivers.

For Citation Platform Partners

The citator that embeds VariLegal owns independent verification.

KeyCite and Shepard's are the authoritative data sources for citation treatment. VariLegal is the independent verification layer that acts on that data. Together they close the gap between AI-generated output and a defensible court filing. Neither does it alone.

The verification gap
CoCounsel tells attorneys to verify. It does not verify for them.
Every major legal AI platform — CoCounsel, Harvey, Lexis AI — carries a disclaimer instructing users to independently verify citations. That disclaimer is an admission. It transfers liability to the attorney and leaves the platform exposed to the question: why didn't you build the verification layer? VariLegal is that layer.
The competitive dynamic
LexisNexis is building the verification narrative around Shepard's.
LexisNexis has deepened its alliance with Harvey. Harvey users now query Lexis AI directly. LexisNexis is positioning Shepard's as the credibility layer in that partnership. The citator that pairs authoritative treatment data with independent adversarial verification owns the defensibility standard for regulated legal AI.
What integration delivers
Every citation verified is one citator lookup at scale.
VariLegal calls the citator API on every citation it processes. Volume-based licensing means VariLegal adoption directly drives citator API revenue. Enterprise law firms demanding documented verification become a joint market rather than a shared risk. The verification layer and the citator data layer grow together.
What compliance velocity delivers
Law firms are asking their AI vendors for documented verification. Now.
EU AI Act enforcement, expanding Rule 11 standing orders, and bar guidance on AI use in practice are converging. Enterprise law firms are not waiting for regulation to land — they are asking their AI vendors today what documented verification looks like. The platform that answers that question first owns the compliance conversation for every firm it serves.

The benchmark is complete. 70% overruled recall gap against KeyCite's own treatment data. 100% fabrication detection. Zero false leniency. Validated against the Stanford RegLab dataset. Document Mode extends verification to private documents — no citator stack required. The work is done — the integration conversation is what remains.

Discuss Integration
Use Cases

Where VariLegal operates.

Litigation

Brief & Memo Verification

Verify AI-generated briefs, memos, and research before filing. Catch fabricated citations before opposing counsel does. Attach the certificate to the matter file as documented due diligence.

Transactional

Contract & SPA Review

Upload the definitive agreement. Run AI-generated contract analysis through Document Mode. VariLegal flags discrepancies in purchase price, dates, thresholds, and operative provisions against the source text.

Compliance

EU AI Act Readiness

Documented human oversight of high-stakes AI outputs, structured, auditable, and produced at the point of use. The record your regulators and clients can read.

Regulated Domains

One verification layer. Multiple regulated outcomes.

Courts sanctioning AI citations. The EEOC scrutinizing algorithmic hiring decisions. Insurance regulators auditing AI underwriting. Energy regulators examining AI-influenced grid operations. These are not separate problems. They are the same regulatory arc: consequential AI decisions, made without independent verification, in domains where the exposure is existential. Legal is the domain where the architecture was built out and measured. The architecture extends to wherever that arc applies.

VariLegal Implemented
AI citation verification prior to court filing. Independent of the generating platform. Three products: citation verification, private document verification, and backward corpus audit.
NDTX · EDMI · Fifth Circuit · Rule 11
Learn more →

The same adversarial verification architecture, Advocate, Adversary, Arbitrator, with a deterministic auditor as System of Record. Regulatory frameworks differ. The methodology does not.

Infrastructure-Grade Compliance
Patent Pending (USPTO) EU AI Pact Pillar II Signatory Implemented System Independent of your AI vendor Built for the post-sanctions judiciary
FederalEEOC / FCRA / ECOA
ColoradoSB 21-169 / CAIA
European UnionEU AI Act
IndustryNAIC / NERC CIP / FDA SaMD

The defensibility layer for AI in regulated environments.

VariLegal was built, benchmarked, and run as a production verification service. The system is currently held offline. Serious evaluation, partnership, and licensing inquiries are welcome, and the environment can be restored for qualified engagements.

Schedule a Technical Conversation

info@getvari.ai

Benchmark methodology and results are published open access on Zenodo.