FIS's new AI banking agent is live. Five months from announcement to commercial deployment is fast. And for mid-market financial services firms watching from the UK, it should prompt an uncomfortable question: if this agent makes a compliance error, who pays? The answer is unambiguous. The regulated firm pays. The FCA, PRA and ICO are clear on this point — outsourcing decision-making to third-party AI does not distribute liability. It concentrates it. Under FCA Consumer Duty PS22/9 and PRA SS1/23, the bank deploying an AI agent remains the decision-maker. The regulator will treat failures as the firm's failures, not Anthropic's or FIS's. This is not a theoretical risk. It is the regulatory baseline.
What FIS's launch reveals is that the industry has moved faster than governance has matured. Banks and fintechs are deploying sophisticated AI agents — systems that make real decisions about customer credit, transaction monitoring, fraud detection, AML screening — without fully resolving how accountability works when the agent fails. The pattern is clear: products like Harvey and Luminance succeeded in legal and document work partly because mistakes are caught before they cause regulatory harm. Banking agents operate in real-time, at scale, in high-stakes environments. The margin for error is zero. Yet we are still treating these systems as experimental. They are not. They are live. And the firms deploying them are learning accountability lessons in real-time, in front of regulators who expect the firm, not the technology vendor, to have engineered fail-safes.
Trovix takes a different view of this problem. We do not believe mid-market regulated firms should deploy third-party AI agents without first building the governance infrastructure to monitor, audit and correct them. That is why we built Trovix Audit — not as a box-ticking compliance tool, but as an operational necessity for firms using external AI systems. Every decision an AI agent makes should be visible, traceable and correctable before it creates compliance liability. Microsoft Copilot and similar general-purpose systems avoid this problem by design — they are assistance tools, not decision-makers. FIS's agent is different. It decides. That difference demands real governance. The gap between FIS's deployment pace and the average UK firm's governance maturity is widening. The FCA will close that gap through enforcement.
What should your firm do now? First: do not assume that using AI from a household-name vendor (Anthropic, OpenAI, Google) reduces your compliance risk. It does not. The vendor's reputation is irrelevant to regulatory accountability. Second: before deploying any AI agent into a decision-making workflow, map how you will monitor, audit and override it. Third: ensure your AI governance framework is documented, tested and demonstrable to regulators. This is not optional under PRA SS1/23. Fourth: if you are considering FIS's agent or similar systems, budget for the governance cost as part of adoption. Many firms underestimate this. It is often larger than the AI tool itself. Finally: audit your current AI deployments now. If you are already using AI agents for compliance-sensitive work — transaction monitoring, AML screening, customer onboarding — and you have not built audit infrastructure around them, you have a liability hole that regulators will find before you do.
Source: Forbes