FairPlay Completes Independent Evaluation of Axle's KYC Screening Agent

LOS ANGELES, Sept. 23, 2026 /PRNewswire/ -- FairPlay, the AI enablement company for financial services, today announced the results of its independent evaluation of Baley, Axle's AI screening agent for sanctions, politically exposed persons (PEPs), and adverse media. Across 1,048 scored test cases, Baley achieved 97.5% accuracy, and FairPlay found the agent fit for its evaluated use across all three screening functions. No material issues were identified.

Banks and payment companies must determine whether a prospective customer appears on a sanctions list, is a politically exposed person, or is associated with adverse media that warrants additional review. Screening providers such as LexisNexis and ComplyAdvantage may return numerous potential matches—particularly for common names—and compliance analysts must determine which can safely be cleared and which require escalation.

That work is far more complicated than matching names. A sanctioned individual may move to another country while retaining the same date of birth. A politically exposed person may legally change her surname without changing the underlying identity or risk being evaluated.

Baley is designed to address these issues. It automates the entity resolution process that compliance analysts traditionally perform manually when reviewing screening alerts, with the goal of reducing the volume of false positives requiring human review. In this case, Baley resolves each candidate record returned by the provider against the customer being screened and returns a verdict of CLEAR or ESCALATE, with a written rationale, so that analysts spend their time on the hits that warrant review.

FairPlay tested Baley using SMEvals, its proprietary suite of subject matter evaluations for agent testing. FairPlay's testing found that Baley performed strongly across all three screening functions, achieving 97.5% accuracy across 1,048 scored test cases. Recall was 97.5%, precision 99.2%, and specificity 97.7%. Baley's performance was consistent across all three screening functions: 97.9% for sanctions, 98.2% for PEP, and 94.9% for adverse media.

FairPlay's SMEvals were designed to test the cases that make real-world screening difficult: non-Latin scripts and transliteration variants, cultural naming conventions, aliases and legal name changes, incomplete or conflicting dates of birth, and recently designated individuals. Because each test case had a known correct answer, FairPlay could measure not only how often Baley reached the right conclusion, but also where it does not, which direction it errs in.

That directional bias determines a bank's true risk exposure. A false positive sends a clean customer into a review queue and costs an analyst time. A false negative clears someone who should have been escalated and can create meaningful compliance risk. FairPlay measured both types of error across each screening function.

FairPlay's accuracy testing was part of a broader independent evaluation that included a review of Baley's design and architecture, an assessment of risks associated with its LLM components, and a review of Axle's governance and change-management practices. FairPlay conducted the evaluation under the principles of SR 26-2, SR 11-7 and SR 21-8, which Axle applies voluntarily to its AI systems.

For banks and fintechs deploying AI in compliance workflows, the central question is straightforward: does handing work once performed by a compliance analyst to an AI agent introduce new risk? FairPlay evaluated Baley as an independent third party with no stake in the outcome.

"AI agents are beginning to perform compliance work that used to sit squarely with human analysts. Banks shouldn't have to take a vendor's word for it that those agents work," said Kareem Saleh, Founder and CEO of FairPlay. "We built SMEvals to answer the question every bank and examiner should ask: how often does the agent get it right, and what happens when it gets it wrong? Baley performed exceptionally well across more than 1,000 test cases."

"Any vendor can tell you their agent is accurate. Very few will hand it to a third party and publish whatever comes back," said Yanni Giannaros, co-founder and CEO of Axle. "We asked FairPlay to test Baley on the cases that actually break screening: transliterated names, aliases, legal name changes, conflicting dates of birth. What our customers care about is not only how often the agent is right, but which way it errs when it is wrong. Those numbers are now on the record, and they can take them to their examiners."

About FairPlay

FairPlay is the AI enablement company for the financial services and insurance industries. FairPlay specializes in testing, tuning, and validating AI systems and agents, helping institutions deploy AI faster, more profitably, and in full compliance. Its products address the toughest challenges in AI deployment, including Bias Testing, Model Optimization, Model and Agentic Validation, Second Look, and Safe AI Sourcing. FairPlay's work has been featured in The Wall Street Journal, USA Today, Politico, American Banker, and Forbes. In 2023, Fast Company recognized FairPlay as a "World Changing Idea." To learn more, please visit fairplay.ai

About Axle

Axle builds AI agents for financial crime compliance. Its agents adjudicate sanctions, PEP and adverse media screening alerts, investigate and disposition transaction monitoring alerts, run business onboarding and enhanced due diligence reviews, and draft SAR narratives, working inside the case management and screening systems institutions already run. Axle's agents are configured from each institution's own procedures and record every decision with its reasoning and supporting evidence. Founded in 2023 and headquartered in San Francisco, Axle is backed by Inovia Capital, Mistral, Uphonest Capital and Diagram, and serves regulated financial institutions across North America, Latin America, Europe, Asia and Africa. Axle is SOC 2 Type II and ISO 27001:2022 certified. To learn more, please visit www.axleruns.com

SOURCE FairPlay