Reported by Mikhail Karataev

(Summary shared below. To read full report, go to: https://www.americanbanker.com/opinion/adding-ai-to-money-laundering-compliance-introduces-new-risks)

Artificial intelligence is quickly becoming a strategic priority for U.S. banks, with many institutions moving to integrate generative AI into business operations, including anti-money-laundering compliance. Industry estimates suggest the technology could improve productivity across banking functions by 20% to 30%. But as banks adopt AI to speed up compliance work, regulators and industry experts are raising concerns about a new class of risks created by systems that can produce convincing but inaccurate information.

The risks are particularly important for financial crime compliance, where investigators rely on accurate information to assess suspicious activity, conduct sanctions reviews and perform customer due diligence. Unlike traditional analytical models, large language models can generate responses that appear professional and technically sound while containing factual errors or fabricated information. In April 2026, the Office of the Comptroller of the Currency, Federal Reserve and Federal Deposit Insurance Corp. clarified that updated interagency guidance on model risk management does not apply to generative AI, recognizing that the technology presents challenges beyond those associated with traditional predictive models.

For AML professionals, the shift could fundamentally change how investigations are conducted. AI can quickly gather information and produce summaries, risk assessments, investigative narratives and legal research. But that efficiency also creates a risk that analysts will rely too heavily on machine-generated conclusions. The role of the investigator may increasingly shift from finding information to validating information and identifying errors in AI-generated analysis. The concern is not simply that AI can hallucinate or produce inaccurate conclusions, but that those mistakes may be embedded in well-written and seemingly credible analysis that is difficult to challenge without careful human review.

The growing use of AI also raises broader questions about accountability. Banking controls have traditionally been built around the principle that significant decisions have an identifiable human owner. With AI increasingly influencing decisions, responsibility may become distributed among analysts who rely on AI outputs, managers who oversee them, governance teams responsible for controls and institutions that determine how the technology is deployed. Banks will therefore need to determine how accountability is assigned when important compliance decisions result from a combination of human judgment and machine-generated intelligence.

AI does not eliminate traditional operational risk; it changes its nature. The biggest concern may not be that machines make mistakes, but that experienced professionals accept those mistakes because they appear credible. For AML and sanctions teams, effective AI adoption will require more than technology controls and model validation. Banks will need governance frameworks that preserve professional skepticism, independent judgment and clear accountability. AI may make financial crime compliance faster and more efficient, but the strength of governance will determine whether that efficiency also makes the banking system more resilient.