Phase 0 findings now available · Phase 1 in development
Frontier LLMs are being deployed against financial crime. We tested whether they're ready.
FinCrimeBench™ is an independent benchmark evaluating large language model performance on Fraud and AML tasks — built and curated by domain experts, anchored to authoritative sources. We test the leading frontier AI providers across expert-authored fraud scenarios and provide benchmarks on the readiness of the LLM models for financial crime prevention and detection. Take a look at what we found and request the report.
Request Research
By submitting your details, you agree to our Privacy Policy.
↓
What we've found so far
We probed four frontier LLM providers across 110 expert-curated prompts spanning select fraud typologies. Performance on conceptual tasks was generally competent. Performance on tasks closest to operational decision-making was not.
Example: Risk Level Assessment
Of the task categories we probed, none matters more operationally than risk level assessment — the judgment that determines whether a case is escalated, a transaction is held, or a restriction is placed on a customer.
The pattern across providers was wrong in the direction that matters most, with confident-sounding rationale.
BEST FRONTIER MODEL
1 in 3
WRONG risk ratings on a four-tier classification. The other three providers ranged from 46%–53% accuracy — meaning the typical frontier model got nearly half of risk ratings wrong.
↓
How FinCrimeBench is built
Prompts curated by practitioners
110 prompts in Phase 0, written and reviewed by financial crime experts with hands-on experience in fraud and AML. Each prompt is paired with an expert answer key reflecting how a competent senior practitioner would respond.
Anchored to recognized standards
The taxonomy and scoring framework draw on guidance from the OCC, FFIEC, FATF, FinCEN, and NIST, with the methodology referenced in the whitepaper.
Provider-neutral and independent
FinCrimeBench is independent research from CoveLabs.ai. While we are in discussions with LLM providers about financial crime training data, FinCrimeBench operates under separate governance with no provider influence over methodology, evaluation, or published results.

↓
Get Involved
Executive Summary
A two-page overview of Phase 0 findings, methodology, and what's coming in Phase 1. Designed to be read in five minutes and shared with your team.
We'll send the summary to your inbox right away.
Full Technical Whitepaper
The complete Phase 0 report — provider-level results, prompt-category breakdowns, scoring methodology, and the detailed findings behind the headline numbers.
We prioritize access for senior leaders accountable for fraud and AML risk at financial institutions, fintechs, and RegTech providers, and LLM researchers.
Executive Briefing
A 30-minute conversation with senior fraud and AML leaders. We walk through findings relevant to your domain — including provider-level results not published in the whitepaper.
We review briefing requests on a rolling basis. Typically confirm or decline within a week.
↓
CoveLabs.ai is leading this independent research effort focused on evaluating and improving how AI systems perform on financial crime tasks. We were founded by industry and AI experts with decades of hands-on experience in fraud and AML, and FinCrimeBench is our first published initiative.
© 2026 CoveLabs.ai LLC · CoveLabs.ai · LinkedIn
FinCrimeBench™ is a proprietary benchmark developed and owned by CoveLabs.ai LLC. The benchmark name, methodology, taxonomy, prompts, expert answer keys, scoring framework, evaluation results, and all associated content, deliverables, and derivative works are the exclusive intellectual property of CoveLabs.ai LLC. The contents of this site and the available whitepaper documents, including all analysis, findings, figures, and text, are protected by copyright. No part of this work may be reproduced, distributed, or used to train, fine-tune, or evaluate machine learning models without the prior written consent of CoveLabs.ai LLC, except for personal, non-commercial reference and for the limited fair-use purposes recognized under applicable law.
Privacy Policy
Last updated: May 2026
1. Overview
Cove Labs ("we", "us", or "our") is committed to protecting your privacy. This Privacy Policy explains how we collect, use, and safeguard information when you visit covelabs.ai or interact with our content. We do not track individuals, sell personal data, or use advertising technologies.
2. Information We Collect
We collect only the information you voluntarily provide through our contact form, including your name, work email address, company, role, and any details you share about your interest or project. We do not collect sensitive personal data, and we do not use automated profiling or tracking of individual users.
3. How We Use Your Information
Information submitted via our contact form is used solely to respond to your inquiry, send requested research materials, and communicate about potential partnerships or services. We will never sell, rent, or share your personal information with third parties for marketing purposes.
4. Cookies & Tracking
We use only essential cookies necessary for the basic functioning of this website. These cookies do not track your browsing behavior, build user profiles, or share data with third parties. We do not use analytics cookies, advertising cookies, or any third-party tracking technologies. No cookie consent banner is required as only strictly necessary cookies are used.
5. Data Retention
We retain contact form submissions only as long as necessary to respond to your inquiry or fulfill the purpose for which it was submitted. You may request deletion of your data at any time by contacting us.
6. Your Rights
Depending on your jurisdiction, you may have rights to access, correct, or delete your personal data. To exercise any of these rights, please contact us at admin@covelabs.ai.
7. Third-Party Form Processor (Tally.so)
Our contact form is powered by Tally.so, a third-party form service. When you submit a contact request, your data (name, work email, company, role, and message) is processed and stored by Tally in accordance with their Privacy Policy (tally.so/privacy). We use Tally solely to receive and manage contact form submissions. Tally does not use your data for advertising or sell it to third parties. We have configured Tally to accept work email addresses only. Personal email domains (Gmail, Yahoo, Hotmail, and similar) are not accepted. You may request deletion of your form submission data by contacting us at admin@covelabs.ai.
8. Third-Party Services
This site may link to external resources. We are not responsible for the privacy practices of third-party websites. We encourage you to review their privacy policies independently.
9. Changes to This Policy
We may update this Privacy Policy from time to time. Any changes will be reflected with an updated date at the top of this page. Continued use of the site constitutes acceptance of the updated policy.
10. Contact
For any privacy-related questions or requests, please contact us at: admin@covelabs.ai

© 2026 CoveLabs.ai LLC