After attending both ACAMS Assembly in Las Vegas and Sibos in Frankfurt, the dominant theme extended beyond AI to Agentic AI. AI and agentic dominated conversations in conference hall sessions and vendor demo booths, and even made its way into regulator panels. Listening to the swirl of conversations on agentic AI, the lines blurred between what’s real, what’s aspirational, and what’s simply misunderstood. Everyone was talking about agentic AI; few could agree what “agentic” should mean, and prompted the question, “What makes AI Agentic?”. Here is what I heard from each group at the conferences on the topics of AI and agentic:
Banks
Risk and compliance leaders were forthcoming about how their institutions are using traditional AI to reduce compliance costs, better detect financial crime, and mitigate risk. Many were clear that they are using traditional AI in in broad, successful production deployments. They are using GenAI in isolated tasks with “humans in the loop”, and talked of steps to mitigate risk and lessons learned in implementing GenAI. For banks, agentic is more roadmap than reality. Executives shared reflections on “this is where we might go next,” rather than “these are the problems we are solving with agentic AI”
Regulators
Regulators are embracing their role in modulating pace of AI adoption and their approaches vary across regions. The result is a three-lane highway of AI adoption:
- United States is in the middle lane. Leaders at the OCC and FDIC have proclaimed an open-minded attitude to using AI to support innovation in compliance. Bank examiners increasingly ask about AI initiatives as part of routine exams, reflecting curiosity and learning rather than immediate scrutiny.
- Europe is in the slow lane. The rate of AI adoption in Europe is being tempered by the complex set of regulations banks face. The documentation and risk assessments required by the AI act raise compliance costs and lengthen development timelines. Fines for breaching GDPR regulations are slowing development of AI models.
- Asia-Pacific is in the passing lane. The Monetary Authority of Singapore provides grants for bank AI projects, while the Hong Kong Monetary Authority fosters experimentation through GenAI sandboxes.
Technology Vendors
While banks keep their agentic AI ambitions close to the vest, vendors are laying their cards on the table. Technology vendors showcased new and updated products with next-generation AI capabilities— graph analytics, machine learning, LLMs. As for agentic AI. several vendors demonstrated fully functional AI systems with multiple agents interacting and driving autonomously towards an outcome. Few of these are generally available but the functionality of the beta versions was impressive. On the other hand, several demoed new “agents.” These demos frequently employed GenAI for ingesting unstructured data and pattern detection, but true agentic capabilities remain aspirational. While some solutions impressed, most limited themselves to single-step automations within preset process flows.
What Makes AI Agentic?
Sorting hype from substance, the agentic bar is high: systems must coordinate complex tasks, proactively plan and adjust strategies, and be able to reach a defined outcome independent of human intervention. Many so-called “agents” just answer questions or use LLMs to perform a single step in a process. In our view agentic AI contains four core characteristics:
- Multi-step, multi-mode: Chains together complex tasks, deploying different AI models as tools for each interim goal.
- Proactivity: Develops strategies independently to reach a target outcome, rather than merely responding to prompts.
- Adaptivity: Alters tactics in real time as new information arises or as conditions change.
- Autonomy: Operates without requiring human intervention, delivering compliance processes and decisions end-to-end.
Looking beyond definitions, the key takeaway from ACAMS and Sibos is that agentic AI is irresistible. Regulatory openness, rapidly evolving vendor capabilities, and stronger AI guardrails are making risk and compliance processes faster, smarter, more autonomous. The jargon might be new, but the momentum is unmistakable. Agentic AI is coming, and this year’s conferences made clear that the industry is preparing—cautiously, collaboratively, and in some places, faster than other will irresistibly improve the efficiency, effectiveness and autonomy of risk and compliance processes.
