Artificial intelligence is transforming anti-money laundering compliance, cutting through transaction volumes and surfacing patterns no human team could catch manually.
But according to Leo RegTech, a firm with two decades in financial regulation behind it, the industry is not being honest about how often AI still gets things wrong.
Leo RegTech recently discussed the point that whilst AI is transforming AML compliance, it should not be making the decisions.
A Microsoft study, DELEGATE-52, tested 19 AI models across 52 professional domains and found frontier models, including top versions of GPT, Gemini and Claude, lost an average of 25% document accuracy across 20 delegated interactions, with degradation reaching 50% overall.
Errors were sparse but severe, compounding silently over longer workflows. A separate Stanford study found legal AI tools from LexisNexis and Thomson Reuters hallucinated between 17% and 33% of the time. Applied to AML files covering PEP status, beneficial ownership and sanctions exposure, such error rates represent significant regulatory liability.
Effective AML compliance depends on firms first building a Business-Wide Risk Assessment and client risk appetite framework under UK Money Laundering Regulations, FinCEN’s Customer Due Diligence rule, or EU AML directives. Any AI tool must ingest that firm-specific context before reviewing client files, rather than relying on generic assumptions.
The AML technology market is projected to reach $9.4bn by 2030, growing at nearly 18% annually. Players named include Quantexa, which has raised $546m and was valued at $2.6bn in March 2025 ahead of a targeted 2026 IPO; Feedzai, which has raised $347m at a roughly $2bn valuation; Napier AI, with around $57m raised; and NICE Actimize, part of NICE Systems.
The legal sector offers a cautionary parallel. In Mata v Avianca, two attorneys were sanctioned after filing a brief containing fabricated ChatGPT-generated cases. In April 2026, Sullivan & Cromwell apologised to a US Bankruptcy Court judge after an emergency motion contained AI-hallucinated citations, admitting its internal review process had failed.
Under the EU AI Act, high-risk provisions covering AI-driven creditworthiness evaluation, originally due 2 August 2026, have been pushed back to 2 December 2027 following May’s Digital Omnibus agreement. Standalone AML transaction monitoring sits outside the high-risk category, though penalties for high-risk breaches could still reach €35m or 7% of global turnover.
Leo RegTech argues the answer is governance, not avoidance: firms need clear AI policies covering permitted tools, human verification requirements, audit documentation and escalation paths. AI should remain an accelerator and first filter, not the final word on client risk classification or SAR decisions.
Read the full Leo RegTech post here.
Copyright © 2026 RegTech Analyst
Copyright © 2018 RegTech Analyst
