AI reshapes AML operations in banks and fintechs

Financial institutions are deploying artificial intelligence in operational anti-money laundering (AML) programmes, moving beyond pilot projects into live systems used for onboarding, know-your-customer (KYC) checks, transaction monitoring and investigations. Banks and fintech firms are applying machine learning, large language models (LLMs) and limited agentic systems to specific tasks such as adverse media screening, alert triage, investigation support and preparation of suspicious activity reports.

Machine learning models are in active use for transaction monitoring and alert scoring. Firms report these models help prioritise alerts by risk level and reduce the volume of manual reviews. LLMs are under trial for document review, extracting relevant passages from customer files and public records, screening news and drafting sections of regulatory reports. Teams use automated summaries from LLMs to populate case-management systems for investigator review.

Onboarding and KYC workflows now often include automated identity verification and risk scoring models that accelerate decisions and flag inconsistent information earlier in the process. In investigations, systems that rank alerts and surface the most relevant documents are used to focus human analysts on higher-probability leads. Some compliance teams deploy automation to assemble the narrative and evidence required for suspicious activity reports while retaining human approval before filing.

Firms are integrating data from transaction engines, sanctions and politically exposed person screening services, customer records and external news feeds to create consolidated risk profiles. Integration work focuses on linking legacy systems with new analytics tools so that behaviour, screening results and media signals can be viewed together during monitoring and investigations.

Governance and regulatory requirements are shaping deployments. Supervisory expectations emphasise transparency, model validation, explainability and documented controls. Institutions are implementing audit trails, version control, model testing and human-review gates to demonstrate how automated outputs are generated and how staff acted on them.

Organisations are choosing different sourcing strategies. Some buy commercial platforms to accelerate deployment and access specialised features. Others build internal teams to retain control over models and data. Many adopt a hybrid approach, using vendor tools for core functions and internal models for unique datasets or jurisdictional compliance needs.

Agentic AI, defined as systems that can sequence actions or make recommendations across multiple tasks, is being piloted in limited investigation workflows. Firms typically restrict agentic systems to support roles and require human investigators to verify recommendations before decisions are made.

Industry conferences and practitioner panels are prioritising case studies that show how institutions move pilots into production while meeting supervisory requirements. Work underway includes data standardisation, documentation of model behaviour and alignment of governance frameworks with regulatory expectations.