重新构想与代理AI的合规性

Enabling Proactive, ContinuousCompliancein a Complex RegulatoryWorld DavidChoiDavidCarreteroAllenMeyerDanielTannebaumFredericThomas-DupuisKim-MaiLe Compliance teams should ambitiously embrace agentic AI systems that arecapable of performing semi-autonomous tasks to reimagine processes, unlockgreater efficiency, and enable proactive, continuous riskmanagement. Compliance leaders today face unprecedented challenges, including diverging regulatoryexpectations, mounting cost pressures, and the urgent need to enhance operationaleffectiveness. AI presents a significant opportunity to reshape how compliance operates,yet our benchmarks and experience indicate that most organizations remain in early stages— experimenting with generative AI in narrow, incremental ways rather than reimaginingworkflows end-to-end. While helpful, these early pilots often fail toachievethe transformative impact and return on investmentanticipated. True transformation requires more than sprinkling AIintotoday’s workflows. The next wave, now emerging amongearlyAI leaders, demands reimagining the compliancefunctionend-to-end: how data is used, how work is divided betweenpeople and machines, and how compliance teamsdelivervalue to business leadership andregulators. AgenticAI Semi-autonomoussystems that can plan,act, and coordinateacross multi-stepworkflows with humanoversight, movingfrom“AI as a tool”to“AI as aco-worker.” Compliance teams must think bigger and embrace agentic AI,semi-autonomous systems (“agents”) that can plan, act, andcoordinate across multi-step workflows with humanoversight. Operating within explicit guardrails, each AI agent is designed to perform a specifictask — such as monitoring for regulatory changes, performing due diligence checks, orpreparing case documentation — and escalate judgment calls to humans. To enable trustand confidence, AI outputs should be transparent, demonstrating data sources, lineage andrationales, and reproducibleruns. Our benchmarks and client engagements have shown that reimagined AI-enabledcompliance functions can enhance output quality and risk detection by up to four times,enable 100% continuous and comprehensive testing and coverage, and reduce time spenton manual, repetitive tasks by 50 to 70%, freeing capacity for higher-value work. AgenticAI enables continuous data collection, analysis, and insight generation across complexcompliance workflows, dramatically enhancing efficiency, accuracy, and scalability. Byflipping the paradigm from “expert-created” to “expert-reviewed,” compliance teams canstreamline operational workflow, empower staff to focus on judgment and strategy, andachieve proactive, continuouscompliance. That said, realizing these benefits will require careful governance, risk management, andplanning. Alongside the significant potential gains, organizations should anticipate newdevelopment and operational costs, such as software build, integration, validation, andmodel maintenance. Agentic AI also introduces new forms of model risk, explainabilitychallenges, and operational dependencies. Finally, achieving the transformation will requirea clear roadmap: defining the ambition and measurable objectives, redesigning human-AI roles, and executing a disciplined rollout supported by strong governance, testing, andworkforceenablement. Agentic AI unlocks continuous, intelligent compliance operationsthat detect and manage risks moreeffectively Agentic AI enables a fundamental shift in how compliance functions operate. AI agentsoperate “always on,” enabling near real-time identification and escalation ofrisks. For example, agentic AI can autonomously track regulatory changes and policy updates,ensuring compliance teams stay ahead of evolving requirements. Similarly, in Know YourCustomer (KYC) and customer due diligence (CDD) processes, agentic AI enables dynamic,ongoing monitoring of customer risk profiles, allowing timely adjustments to due diligencebased on newdata. The art of the possible extends even further. In the long run, AI has the potential to upendthe entire paradigm of compliance monitoring. Today, banks rely on rules-based alertsbecause transaction volumes are too great to review exhaustively. With near-unlimited AIcapacity, firms could one day analyze 100% of transactions in real time — not just flaggingissues after the fact but stopping risky transactions before theyoccur. How the compliance workforce will change in thefuture Today, most compliance functions remain headcount-heavy and operationally skewed.Highly trained staff spend a small share of their time on specialized work such asgovernance, advisory, and risk assessments, compared with execution tasks such asinvestigating cases, testing controls, reconciling data, and processing regulatory changes.These activities are essential for regulatory compliance but do little to elevate the strategicvalue of compliance and take away time that could better be spent on riskdecisioning. The future looks radic