FinProof v1 is an open adversarial benchmark designed to test AI guardrail systems in banking, financial services, and insurance. It covers 7 attack categories across professional and retail conversational registers and was published by Zytra under a CC BY 4.0 license. The dataset was last updated on Hugging Face in May 2026.
Use Cases
- Stress-testing AI guardrail robustness based on the 7 specified attack categories.
- Evaluating model performance across professional and retail conversational registers mentioned in the description.
- Benchmarking distribution coverage using quantum-augmented generation techniques referenced in the description.
Strengths
- Covers 7 distinct attack categories for systematic evaluation.
- Includes both professional and retail conversational registers.
- Uses quantum-augmented generation for distribution coverage.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count and dataset size are unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Zytra via Hugging Face.
- Collection Method
- Likely generated or curated for adversarial testing.
- Time Range
- Published in 2026.
- Freshness
- Last updated 2026-05-09 06:59:27; freshness should be verified.
- Geography
- Coverage is not specified.