FinProof v1.0 — 5,389 prompts · 22 BFSI domains

The First Benchmark Built for GenAI Guardrails in Financial Services

FinProof is a purpose-built benchmark dataset for evaluating GenAI guardrail performance in regulated financial institutions. Designed by practitioners. Mapped to RBI, EU AI Act, SR 11-7, ISO 42001, and India DPDP Act 2023.

Open Dataset — HuggingFace
Zero Training Contamination

Industry-Leading Benchmark

FinProof by the numbers — comprehensive coverage, rigorous evaluation, open access

5,389
Evaluation Prompts
Adversarial test cases across BFSI
22
BFSI Domains
B-01 through B-11 · D-01 through D-11
0%
Training Contamination
Verified zero overlap with public LLM data
+46.7
F1 Points
Over nearest competitor (Granite Guardian)
Features

Built for GenAI Safety & Compliance

Built by financial AI researchers for financial AI researchers. FinProof provides the tools you need to ensure your models are production-ready.

21 Guardrail Actions Tested

Covers the complete taxonomy of GenAI guardrail actions across Input, Processing, and Output layers — from Block and Redact to Agent Handoff Validation and Rollback.

Regulatory Mapping

Every domain and prompt mapped to specific obligations under EU AI Act, SR 11-7, ISO/IEC 42001, and India DPDP Act 2023.

QCBM Adversarial Generation

Prompts generated using Quantum Circuit Born Machines — 89,022 hashes evaluated, zero training contamination confirmed. Novel adversarial patterns no static classifier has seen.

Four-Tier Access Model

Tier 1: Public (JSONL download). Tier 2: Email gate (extended set). Tier 3: Research agreement (full corpus). Tier 4: Withheld test set with public SHA-256 integrity hash.

Plain JSONL — Run In-Perimeter

No external API calls required. Institutions with restricted outbound access can run the full evaluation harness entirely within their own infrastructure.

Open Evaluation Harness

CLI tool available on GitHub for running the test suite against any guardrail stack. Score your own guardrail implementation against the FINPROOF standard.

Use Cases

Benchmark Every BFSI Application

From credit risk to fraud detection, FinProof covers the full spectrum of financial AI applications

Credit Risk Assessment

Evaluate credit scoring models against diverse borrower profiles with ground-truth outcomes spanning 10+ years of credit history.

Multi-factor credit scoring validation
Demographic fairness testing
Economic cycle stress testing
Regulatory compliance checks
94.7%
Default Prediction Accuracy
100%
Bias Detection Coverage
2.5M+
Test Scenarios
Leaderboard

FINPROOF Guardian Model Leaderboard — May 2026

B-01..B-07 BFSI domain evaluation · 5 models · BFSI macro F1 · Precision · Recall · FPR

#ModelBFSI F1
REF
Lynx v1.4Reference Baseline
Zytra Tech Solutions · 184M params · DeBERTa-v3
0.977
Granite Guardian 3.3
IBM Research · 8B params
0.813
ShieldGemma 9B
Google · 9B params
0.731
LlamaGuard 3
Meta AI · 8B params
0.569
4
WildGuard 7B
Allen AI · 7B params
0.346

Per-Domain F1 — B-01 to B-07

★ denotes best score per domain

DomainLynx
B-01Investment advice0.985
B-02KYC / Card fraud0.971
B-03Employment fraud0.984
B-04Regulatory hallucination0.926
B-05Predatory lending1.000
B-06Insurance conduct0.976
B-07Financial instruments1.000

Official scores computed on withheld Tier 4 test set. Reference baseline shown for context.

Submit your guardian model: finproof@zytratechnologies.com

Access Tiers

Dataset Access Tiers

FinProof uses a structured access model to protect evaluation integrity while enabling broad research access.

Tier 1 — Public

Open access. No registration required.

Free
  • 1,078 prompts (JSONL format)
  • 8 representative BFSI domains
  • Evaluation harness (GitHub)
  • Full regulatory mapping documentation
Most Popular

Tier 2 — Extended

Register with institutional email to access the extended set.

Email Gate
  • 1,347 prompts
  • 14 BFSI domains
  • All Tier 1 content
  • Benchmark scoring scripts
  • Priority release notes

Tier 3 — Full Corpus

Full dataset under a signed research use agreement.

Research Agreement
  • 1,886 prompts
  • All 22 BFSI domains
  • Detailed annotation metadata
  • Regulatory mapping per prompt
  • Academic citation support

Tier 4 — Withheld Test Set

1,078 hard adversarial prompts held back from all public releases.

Evaluation Only
  • Never released publicly
  • SHA-256 integrity hash published
  • Used for official leaderboard submissions
  • Prevents benchmark overfitting

All tiers include access to documentation, community support, and regulatory mapping. Tier 4 is used exclusively for official model evaluation submissions.

Ready to Benchmark Your Financial AI?

Join 200+ leading financial institutions using FinProof to validate their AI models. Get started with our free research tier or contact us for enterprise solutions.

No credit card requiredFree academic accessSOC 2 compliant