Compliance Teams Cross the Adoption Threshold on AI Agents, Industry Survey Finds

A comprehensive practitioner survey released in mid-July by identity and financial-crime research firm Liminal supplies the clearest quantitative evidence to date that artificial intelligence has moved decisively past the experimentation phase within anti-money laundering compliance functions and into mainstream operational deployment. Ninety-five percent of surveyed practitioners report that they are either using or actively planning to use AI agents within compliance workflows, with adoption highest in two of the functions most central to a modern AML program: eighty-three percent in know-your-customer processes and seventy-eight percent in transaction monitoring. Those figures represent a meaningful shift from the more tentative, pilot-stage adoption rates that characterized industry surveys as recently as 2024, and they suggest that the budget pressure, false-positive fatigue, and increasingly receptive regulatory posture that have been building for several years have now converged into a genuine tipping point.

The survey’s most operationally useful finding concerns where institutions are choosing to deploy AI agents first, and the pattern is notably consistent across respondents: seventy-nine percent of KYC teams and seventy-six percent of transaction monitoring teams identified sanctions and PEP screening as their initial AI agent deployment. That convergence reflects a deliberate risk-management calculus rather than coincidence. Sanctions and PEP screening are high-volume, rule-based tasks with clearly defined match criteria and outcomes that remain straightforward to audit after the fact, making the category the natural entry point for institutions seeking to demonstrate AI reliability to skeptical examiners before extending automation into judgment-intensive areas like suspicious activity narrative drafting or complex typology detection, where the consequences of an unexplainable false negative are considerably harder to defend during an examination.

That caution around explainability is well founded, and the survey’s data on regulatory rejection is arguably its most important finding for compliance leadership weighing further AI investment. Among practitioners whose AI-based AML systems have faced regulatory pushback, seventy-two percent cited potential bias as the top concern examiners raised, sixty-seven percent cited a lack of explainability, and fifty-six percent cited concerns about raw accuracy. Critically, the survey found that regulators are not rejecting AI-based systems categorically; rather, approval correlates strongly with whether the AI operates inside a governed workflow featuring audit trails, formal model governance documentation, and evidence of real examiner engagement with how the system reaches its conclusions. Institutions that have already run AI systems in live production, with that governance infrastructure in place, hold what the survey characterizes as a fifty-two-point approval advantage in examinations over institutions still in the piloting stage — a gap large enough to function as a genuine competitive and regulatory differentiator, not merely an efficiency one.

The expected operational payoff from AI agent deployment, according to surveyed transaction monitoring practitioners, centers overwhelmingly on alert triage rather than novel detection capability: seventy-one percent expect AI agents to accelerate alert resolution and reduce investigation backlogs, the single most commonly cited benefit, ahead of the sixty-three percent who expect improved detection of complex laundering patterns and the sixty-one percent anticipating reduced manual workload. That ordering is instructive for institutions building an internal business case for AI investment: the near-term, most defensible return on investment remains operational efficiency in a function most compliance leaders already know is understaffed relative to alert volume, rather than a speculative claim about superior detection of previously invisible laundering typologies, a claim regulators have historically scrutinized far more skeptically.

For compliance officers assessing where their own institution sits relative to this survey’s findings, the practical takeaway is less about whether to adopt AI — the ninety-five percent adoption or intent-to-adopt figure suggests that question is functionally settled within the industry — and more about sequencing and governance discipline. Institutions still debating AI adoption in the abstract are, per this data, now a clear minority position; the more consequential open question for the majority already deploying these tools is whether their governance infrastructure — audit trails, bias testing, and documented human oversight — is mature enough to survive the examiner scrutiny that the survey’s own data shows is now the primary determinant of whether an AI-based AML system earns regulatory approval or triggers a formal finding.