2025年金融犯罪预防的实用应用：AI采用指南

Contents IntroductionTypical AML processUnderstanding generative AIUnderstanding predictive AIUnderstanding agentic AIAI-driven AMLDetect new and hidden riskAlert prioritizationStreamlining investigations and workflowsEnhancing investigation processesAutomate complex workflows with agentic AIHow to start your AI adoptionSymphonyAI solutions Introduction Financial crime prevention is changing. With an influx of new technology, criminals arefinding it easier than ever to commit moneylaundering, fraud, and sanctions evasions. The tools available to criminals – such as usingAI – are also available to financial institutions,enabling them to detect and prevent financialcrimes. Tech spend is the priority for 69% of banks andmore than 80% are scoping or engaging in AIinitiatives in financial crime. Despite this, just 46% of banks reported to theBank of England that they have only a ‘partialunderstanding’ of the AI technologies they use. Though parameters and cautions remain, thereis increasing acceptance by regulators that usingAI can help to mitigate crime. This guide aims to help you understand thepractical applications of AI in AML processes. Sources: Chartis analysis 2024 | Bank of England and FCA report, 2025. Understanding generative AI applications within FinCrime, such as providinggreater contextual explainability for detection,investigation, and reporting activities. With theability to ask questions and interrogate data,investigators can use generative AI copilots toassist in every aspect of their investigations. Generative AI is one of the more commonapplications of AI in use today. It uses naturallanguage processing models and prompts toproduce text, images, or analysis. By makingcomplex data accessible and understandable,generative AI has many A generative AI copilot can search documentsand populate and summarize key findings in afraction of the time it would takes a human. When paired with predictive AI, the results areeven more impactful. Understanding predictive AI While generative AI use in financial crimeprevention is in the foreground, predictive AIcan work at speed in the background.Predictive AI uses machine learning andadvanced algorithms to analyze vast amountsof data extremely quickly. It identifies complex institution’s risk management strategy and riskappetite. When predictive AI works alongsidegenerative AI, the outcome is a modern,seamless, and comprehensive approach tofinancial crime prevention. patterns hidden within transaction data to makepredictions about the level of risk associatedwith suspicious or anomalous behaviors.Predictive AI models continually learn andimprove via risk assessment feedback and canbe configured to align with each financial Understanding agentic AI Agentic AI, and the use of so-called AI agents,is perhaps the most exciting development infinancial crime prevention today, significantlystrengthening the industry'sdefenseagainstemerging and evolving threats. AI agents areautonomous decision-making 'bots' that use adaptive learning to complete tasks. Theseagents all have set roles, such as researchingnames or analyzing transactions, and can workwith one another to come to conclusions that areonly then interrogated by an investigator. Able towork in real-time, identify suspicious activity and patterns, they also learn from historical cases,evolving as they absorb more data. Seen by mostas the future of financial crime prevention, theyoffer an exciting glimpse at where the industry isheading, with some AI agents already available toimprove working processes. AI-driven AML An AI-driven transaction monitoring process builds on the traditionalapproach and enhances it for modern, global finance. Rules-based detection engines can be augmented with predictive AImodels, enhancing their accuracy and exposing emerging risk vectorsthat traditional rules are unable to detect. When rules are triggered, predictive AI assesses the risk level andprioritizes alerts, identifying false positives and ensuring alertsrepresenting the greatest risk are reviewed first. Predictive AI models continue to learn over time via risk assessmentfeedback, enabling ongoing improvements to better identify suspiciousbehavior and hidden risk typologies. A generative AI powered copilot can analyze vast amounts of disparatedata, assisting the human investigator by providing relevant alertcontext and helping to answer investigator questions. Suspicious Activity Reports (SARs) can also be generated by a copilotin a fraction of the time, ensuring all relevant investigation information isincluded and allowing SARs to be far more consistent. Let's examine some of the applications in more detail. Detecting new and hidden risk Use case How it helps In transaction monitoring, predictive AI modelsexpose suspicious behavior and transactionanomalies that are easily missed by rules alone. Detection of illicit customer activity is the firstline of defence in helping to prevent finan