Agentic AI
Also called AI agents, autonomous AI agents
Agentic AI is an AI system that pursues a goal by planning steps, using tools such as databases or other software, and acting with limited human input. It is usually built on a large language model. In compliance, agents gather evidence for alerts, draft case notes and run checks that once needed an analyst.
How it works
An agent receives a goal, such as reviewing an AML alert. It breaks the task into steps, queries internal systems and external sources, reads documents and records its findings. It checks results and repeats until it reaches an answer or hands over to a person. Firms set permissions so agents can read data but cannot close cases or move money without approval.
Agents raise new control questions. They act across several systems, so errors can spread faster than with a single model. Firms need logs of every action, clear limits on authority and named humans accountable for outcomes. Singapore's IMDA published a Model AI Governance Framework for Agentic AI in January 2026. EU AI Act duties apply where an agent is used for a high-risk purpose.
The rules
Guidance on bounding agent risks, keeping humans meaningfully accountable, technical controls and end-user responsibility.
Organisations deploying agentic AI, built in-house or boughtApplies risk-based duties by use case, so an agent used for credit scoring or recruitment carries high-risk obligations.
Providers and deployers of high-risk AI systems, including agents used for Annex III purposesCommon questions
How is agentic AI different from generative AI?
Generative AI produces content such as text or images in response to a prompt. Agentic AI uses that capability to take actions. It plans steps, calls tools, reads results and continues until a goal is met, with less human direction at each step.
Where do compliance teams use AI agents?
Common uses include gathering evidence for transaction monitoring alerts, adverse media research, KYC file reviews, drafting suspicious activity narratives and mapping new regulations to internal policies. Most deployments keep a human reviewer before any final decision.
Are there rules specific to agentic AI?
Few rules name agents directly. Singapore's IMDA published a governance framework for agentic AI in January 2026. Elsewhere, existing AI, data protection, outsourcing and senior manager accountability rules apply to how agents are used.
