Scott Zoldi, Chief Analytics Officer, FICO.
As many banks continue their struggle to close crucial AI governance gaps, there’s a new risk on the horizon: giving AI agents broad autonomy over critical business processes. Even as agentic AI rides the crest of the hype cycle, Gartner, Inc. warned in April 2026 that fully autonomous agents were “not ready for the majority of enterprise use cases.”
The risk of ungoverned agentic AI use is real and growing. Despite this, Gartner further reported that more than 60% of organizations were expected to deploy AI agents by 2028, up from 17% that had done so at the time.
The Immediacy Of AI Agent Governance And Oversight
While an agentic disaster may not yet be upon us, the trajectory is clear: The downsides of AI, such as lack of interpretability, hallucinations and sycophancy, could easily wreak havoc if amplified through multiple AI agents and left unchecked.
Moreover, the situation becomes even more complex––and the risk stakes higher––when agents are allowed to self-adapt. The interpretability and auditability of agents’ decisions can be complicated by factors ranging from operational sensitivities to environmental conditions at the time of execution, often with material customer impact.
Why Blockchain-Based Governance Is Optimal
A plethora of agentic AI governance options is becoming available in response. They run the gamut from agent verification platforms to tools created to prevent improper privilege escalation and avert rogue agentic behavior. These methods may address agent security, but they lack fundamental understanding and control of collective agent behaviors and audit decisions––an absolute must in business environments.
For enterprises seeking to operationalize agentic AI, blockchain technology can effectively govern AI agents at both their point of creation and in production.
FICO’s patented and patent-pending blockchain-based AI governance approach starts at the beginning, codifying the development of an AI agent to ensure its behaviors can be explained, monitored, controlled and audited. In production, agentic AI task blockchains capture the interdependency of multiple agents and their blockchain governance to provide interpretability of agentic AI decisions produced by a combination of agents.
The ‘Secret’ Of Agentic AI Governance
The key to building effective, governable agentic AI systems draws on a quote that many have attributed to author Mark Twain: “The secret of getting ahead is getting started. The secret of getting started is breaking your complex overwhelming tasks into small manageable tasks, and starting on the first one.”
That same logic underpins the future of agentic AI. Effective, governable systems are built by breaking complex tasks into small, manageable ones. In practice, this approach has produced two kinds of interpretable and auditable GenAI models that serve as the building blocks for AI agents.
The first, focused language models (FLM), accompanied by trust scores, enables a sharp focus on both the domain and the task of language-based tasks. The second type of GenAI model, the focused sequence model (FSM), addresses real-time transaction decisioning. FSMs analyze a customer’s transaction history all at once to determine if a current transaction is fraudulent or risky.
The Many Components Of A Single Decision
In an anti-money laundering (AML) deployment at a bank, the decision of whether an incoming deposit may be terrorist-financed, laundered or otherwise illicit can be broken down into dozens of components. An agent executes each one and is focused on that single, specific task. In turn, these agents’ micro-decisions are orchestrated in milliseconds into “yes,” “no” or “flagged for human review” decisions.
In this example, the first batch of FLM agents answers the fundamental AML question: “Who is the customer making the deposit?” Multiple agents execute the bank’s perpetual know your customer (KYC) processes, assessing whether the customer is who they say they are and if they are or aren’t present on voluminous sanctions lists.
FSM agents are simultaneously tasked with determining whether this type of deposit may likely be illicit. Individual agents will consider the customer’s transaction history to assess multiple facets of the deposit, including the amount and method of deposit as well as the account, institution and country where the deposit has been drawn.
A complex decision (“Is this deposit transaction compliant with AML regulations or not?”) can be orchestrated by using multiple FLM and FSM agents to solve smaller, simpler tasks. Cumulatively, these tasks reach a final decision on whether to refer the transaction for investigation of potential money laundering.
Blockchain Building Control Into Agentic AI
Blockchain governance is foundational to enabling trust in enterprise-class agents and agentic AI systems. In the example above (and in any agentic AI deployment), the development of each individual FLM, FSM and any other AI agent is persisted to its own development blockchain. This includes the permissible interactions of each agent with others and how to monitor for allowed use of outcomes, which are also codified in the blockchain.
Orchestration of the resulting agentic workflows is controlled by an additional agentic AI task blockchain that contains auditable details to govern any customer decision. These details, or “decision DNA,” include the data used for decisioning, such as the agents’ decisions, monitoring and internal adaptive state as well as the interdependencies of various agents at the moment of decision—all of which impact critical customer outcomes.
A fabric of intricately linked blockchains may appear complicated, particularly when compared to the “point and click” agentic orchestration some providers claim. In my mind, however, accountability isn’t optional in today’s AI-driven world.
Just as relational databases eclipsed flat file systems in the 1970s, I believe two advancements are inevitable: Blockchain-based governance will become the gold standard for keeping AI agents running as intended, and decision DNA for agentic AI will become the analog of explainable AI for traditional analytic models.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
