Bretton AI’s KYC, AML, and sanctions AI agents have completed an independent model validation by accounting and advisory firm Kaufman Rossin.
Bretton AI, which provides AI-native operations for financial institutions’ back-office workflows, has announced that its know-your-customer (KYC), anti-money laundering (AML), and sanctions AI agents have completed an independent model validation by US accounting and advisory firm Kaufman Rossin. The review examined the governance, processing, outputs, and conceptual soundness of the agents, including model accuracy and stability, explainability and decision traceability, change management, ongoing monitoring, and model risk governance. Kaufman Rossin also assessed Bretton’s approach against industry practices and regulatory guidance.
Combining LLM reasoning with controls
Jason Chorlins, Principal at Kaufman Rossin, stated that Bretton’s combination of investigative reasoning by large language models with deterministic rule-based logic, structured output formats, and mandatory human review before final decisions balances the analytical capabilities of these models with the transparency and control required in regulated compliance. According to Chorlins, independent model validation provides an important assessment of the controls and processes supporting the use of such agents in regulated environments.
Beyond third-party review, Bretton’s approach includes what it calls Trust Infrastructure, which combines model risk management, policy controls, decision traceability, and continuous evaluation. Rather than relying solely on general-purpose AI benchmarks, Bretton builds evaluations around the specific tasks its agents perform and the policies of the institutions using them, checking that tasks are carried out correctly and consistently. Feedback from previous investigations informs improvements, and the framework is used to assess new capabilities as underlying AI models change.
Oversight for AI in banking
Will Lawrence, CEO and Co-Founder of Bretton AI, stated that the standard for AI inside a bank is very different from that for a general-purpose AI tool. According to Lawrence, when an agent investigates financial crime or reviews a customer for onboarding, reaching the right answer is only part of the job, as banks also need to know that the agent followed their policies, understand how it reached its decision, and be able to evaluate its performance over time. He added that independent validation provides scrutiny beyond Bretton’s own evaluations.
Bretton works with financial institutions including Robinhood, Mercury, Gusto, Lead Bank, and Coastal Community Bank. The validation applies to AI agents a type of independent model review that banks typically use for their risk and compliance models, as institutions consider how far to rely on AI in regulated back-office work.
