3 USPTO Provisional Patents

Adversarial AI Verification for Regulated Industries

Three AI agents debate every decision before it matters. The reasoning becomes the audit trail. Humans stay in control.

See Live Deployments
The Architecture

Three Agents Debate. Audit Trail Generated.

Every high-stakes determination runs through adversarial verification before action. Not a score. A complete reasoning chain that shows why the conclusion would survive scrutiny.

Advocate
Builds the case
+
Adversary
Challenges for gaps
+
Arbitrator
Resolves the debate
=
Audit Trail
Cryptographic record

AI doesn't just decide. AI defends its reasoning before the decision matters.

Live Deployments

One Architecture. Multiple Regulated Domains.

VARI's adversarial verification is domain-agnostic. The same patent-protected methodology applies wherever AI decisions must withstand regulatory scrutiny.

Legal AI Verification

VARI-LEGAL

Adversarial verification of AI-generated legal research and citations

  • 100% hallucination detection on Stanford RegLab cases
  • California SB 574 citation verification compliance
  • Deterministic scoring with tamper-evident audit trails
View Pricing
Insurance Compliance

VARI-INSURE

Adversarial verification for AI-generated insurance decisions

  • Colorado SB 21-169 AI discrimination compliance
  • Prohibited factor and proxy variable detection
  • FCRA, ECOA, and rate filing verification
View Pricing
Healthcare Compliance

VARI-MED

Adversarial verification for healthcare AI coverage decisions

  • TRAIGA adversarial testing safe harbor
  • Prior authorization AI bias detection
  • NIST AI RMF compliance mapping
View Pricing
Energy Compliance

VARI-NERC

Adversarial verification for NERC CIP compliance

  • CIP-007-6 patch management verification
  • Catches gaps before auditors do
  • 15-second parallel verification
View Pricing
Hiring Compliance

VARI-HIRE

AI-powered recruiting with human oversight architecture

  • EU AI Act Article 14 compliant
  • Colorado AI Act SB 24-205 ready
  • Bias detection before human review
View Pricing

Domain-agnostic by design. The December 2025 provisional explicitly covers "regulated decision domains including but not limited to employment, financial services, healthcare, and critical infrastructure." Five domains now live.

Compliance Coverage

Built for Regulations That Matter

VARI satisfies compliance requirements architecturally, not procedurally. Fines up to $1.6M per violation per day.

Legal AI Verification

California
CA SB 574
Nondelegable Citation Verification
Attorneys must personally review all AI-generated legal work and verify citations. Cannot delegate decision-making to AI tools. VARI-LEGAL provides the independent adversarial verification layer that satisfies the statutory review requirement with documented audit trails.
Live in VARI-LEGAL
Citation Verification
Fabricated Citation Detection
Three-agent adversarial review with primary source retrieval from CourtListener. Catches fabricated cases, mischaracterized holdings, and overruled standards. 100% detection on Stanford RegLab hallucination test cases.
Live in VARI-LEGAL
Deterministic Scoring
Auditable Verdict Classification
Verdicts calculated from measurable signals: citations found, challenges sustained, fabrication markers detected. Same input produces same score every time. SHA-256 tamper-evident audit trails for every verification.
Live in VARI-LEGAL
Multi-Source Verification
AI-Agnostic Output Verification
Verifies output from any legal AI system: CoCounsel, Lexis+ AI, ChatGPT, Gemini. Distinguishes fabricated citations, mischaracterized holdings, overruled standards, and incomplete-but-accurate summaries.
Live in VARI-LEGAL

Insurance AI Compliance

Colorado
CO SB 21-169 § 10-3-1104.9
AI Discrimination in Insurance
Insurers must demonstrate AI-driven underwriting, pricing, and claims decisions are not unfairly discriminatory. VARI-INSURE detects prohibited factors, proxy variables, and disparate impact risk with deterministic scoring and audit trails.
Live in VARI-INSURE
Federal
FCRA 15 U.S.C. § 1681m
Adverse Action Notice Requirements
AI-generated denial letters must include CRA identification, statement that CRA did not make the decision, right to free copy within 60 days, and right to dispute accuracy. VARI-INSURE auto-fails incomplete disclosures.
Live in VARI-INSURE
NAIC Model Bulletin
AI Governance in Insurance
NAIC requires insurers to establish governance frameworks for AI use in underwriting and claims. VARI-INSURE provides the adversarial testing layer with complete audit trails satisfying documentation requirements.
Live in VARI-INSURE
Unfair Claims Settlement Practices
AI Claims Adjudication Verification
AI-driven claims decisions must conduct reasonable investigation before denial and provide prompt fair settlement when liability is clear. VARI-INSURE verifies actuarial justification, independent inspection records, and settlement adequacy.
Live in VARI-INSURE

NERC Critical Infrastructure Protection

CIP-007-6
Systems Security Management
Most violated NERC standard. Patch management with 35-day evaluation cycles. Port and service monitoring.
Live in VARI-NERC
CIP-010-4
Configuration Change Management
Second most violated standard. Baseline configurations, change authorization, vulnerability assessments every 15 months.
Live in VARI-NERC
CIP-015-1
Internal Network Security Monitoring
Multi-tool ingestion layer normalizes alerts from Dragos, Nozomi, and Industrial Defender into ESP-scoped events. Adversarial verification loop evaluates R1 detection and evaluation evidence; audit trail covers R2 retention and R3 data protection.
Live in VARI-NERC
CIP-004-6
Personnel & Training
Background checks, security awareness training, access management and revocation procedures.
Live in VARI-NERC

Healthcare AI Compliance

Texas
TRAIGA § 552.105(e)(2)
Adversarial Testing Safe Harbor
First US law to explicitly recognize adversarial testing as creating an affirmative defense. Penalties of $10K-$200K per violation. VARI's three-agent architecture maps directly to the statutory requirement for continuous adversarial verification with NIST AI RMF substantial compliance.
Live in VARI-MED
California
CA SB 1120
Licensed Physician Review of AI Denials
A licensed physician must review all AI-driven utilization review decisions resulting in denial. VARI's adversarial loop verifies genuine clinical judgment vs. rubber-stamping by analyzing review time, volume, patient record access, and override patterns.
Live in VARI-MED
Texas
TX SB 1188
Physician Review of AI-Generated Records
Practitioners using AI must review all AI-generated clinical content consistent with Texas Medical Board standards. VARI verifies attestation documentation, review thoroughness, and US-based storage requirements.
Live in VARI-MED
Arizona
AZ HB 2394
AI Cannot Be Sole Basis for Denial
A qualified healthcare professional must make the final determination. AI cannot be the sole input to a denial decision. VARI verifies that individual clinical data was reviewed and independent professional judgment was exercised.
Live in VARI-MED
Federal
CMS-0057-F (42 CFR § 438.210)
Prior Authorization Decision Transparency
Plans must provide specific reasons for PA denials including clinical rationale and criteria applied. VARI verifies that denial notices contain specific clinical reasoning, guideline citations, AI role disclosure, and appeal rights.
Live in VARI-MED

AI Hiring Compliance

United States
California
CPPA ADMT Regulations
Pre-Deployment & Opt-Out Rights
All four requirements (candidate notice, opt-out routing to human review, plain-language explainability, and automated risk assessments) are handled by dedicated VARI modules. Structurally prevents replacing human decision-making.
Satisfied by architecture
Colorado
SB 24-205 AI Act
Algorithmic Discrimination Prevention
Every requirement maps directly to a VARI mechanism. The adversarial loop covers continuous risk management, the bias agent handles live discrimination detection, and correction and appeal workflows are embedded, not optional add-ons.
Satisfied by architecture
Federal
Title VII + EEOC Guidance
Disparate Impact & Vendor Liability
Three requirements, three clean answers. Bias detection is pre-decisional, not triggered by complaints. The audit trail is built for EEOC investigation readiness. Human decision authority is preserved at every stage.
Satisfied by architecture
Europe
EU AI Act
Article 14 Human Oversight
High-risk AI systems require human oversight enabling understanding of limitations, automation bias recognition, and override capability.
Satisfied by architecture
EU GDPR
Article 22 Automated Decisions
Right not to be subject to purely automated decisions with legal effects. Fines up to €20M or 4% of global revenue.
Satisfied by architecture
Canada
Québec
Law 25 Section 12.1
Automated Decision Prohibition
Human review is structural, not symbolic. Explainability artifacts are generated per decision, PIA documentation is produced at deployment, and access/correction workflows are built into the governance layer. Zero conditional language.
Satisfied by architecture
Ontario
Bill 149 Working for Workers
AI Disclosure in Job Postings
Requirements are narrow and binary: disclose AI use, enforce employer-size thresholds. VARI handles both automatically: disclosure text is injected into postings, jurisdiction logic enforces size rules. No configuration gaps.
Satisfied by architecture
Federal
PIPEDA
Consent, Access & Accountability
The most compact requirement set in the hiring compliance landscape. Consent is captured and logged. Decision traceability is complete. Named accountability artifacts are generated for compliance officers. No conditionals.
Satisfied by architecture
The Difference

Documentation vs. Verification

Without VARI
Compliance Documentation
AI collects evidence and organizes it. When an auditor asks "why does this satisfy the requirement?" - there's no answer. The documentation shows what was done, not why it's defensible.
With VARI
Compliance Verification
Before any determination is finalized, three AI agents argue about whether it would survive scrutiny. The debate becomes the audit trail. Gaps are caught before auditors find them.

Explore VARI Products

Pricing, features, and subscriptions for all five VARI domains are available at our product hub.

Visit getvari.ai