AI agents face regulatory hurdles as finance demands KYC compliance

Developers describe AI agents as loops that call large language models (LLMs) with tool schemas, allowing the model to request external actions and iterate until a final answer is produced. This architecture differs from plain stateless LLM calls by adding a tool‑calling layer that can execute functions such as fetching metrics or handling weather queries.

Financial systems, however, impose strict pre‑conditions before any transaction can be executed. Four key checks must be satisfied: (1) authentication that a real human is present, often enforced by Strong Customer Authentication in the EU; (2) authorization limits defined by mandates or power‑of‑attorney documents; (3) verification of the actor’s identity and cleanliness through Know‑Your‑Customer (KYC) and anti‑money‑laundering (AML) processes; and (4) compliance with regulator‑mandated limits. An AI agent that aims to spend a user’s money must therefore be able to answer these questions, a capability that current agents lack. The gap highlights the need for robust identity and trust mechanisms before AI representatives can operate in real‑world finance.