A chorus of senior financial regulators is growing louder around one of the most consequential emerging threats to global financial stability: agentic artificial intelligence. Unlike conventional AI tools that assist human decision-making, agentic AI systems act autonomously — executing trades, managing portfolios, and navigating complex financial processes without moment-to-moment human oversight. Central bankers and watchdog chiefs are now warning that existing regulatory frameworks were simply not designed to govern machines that can act, adapt, and transact at speeds and scales no human institution can match.
Financial Conduct Authority Chief Executive Nikhil Rathi has emerged as one of the most prominent voices in this debate, articulating a regulatory philosophy that departs sharply from the blunt-instrument approach regulators have historically applied to emerging technology risks. "We need to think about new tools and a different way of working with the [AI] market in a more collaborative way," Rathi said, signalling that the United Kingdom's primary markets and conduct regulator sees coercive prohibition as an insufficient — and potentially counterproductive — response to the agentic AI challenge.
The significance of that framing should not be understated. Rathi's call for collaboration rather than confrontation reflects a growing recognition among sophisticated regulatory bodies that agentic AI cannot simply be regulated out of existence. The technology is already embedded in trading infrastructure, credit assessment pipelines, fraud detection systems, and increasingly in autonomous financial advisory services. Attempting to impose static, prescriptive rules on systems that learn and evolve in real time risks both stifling legitimate innovation and creating a compliance theatre that provides no genuine protection against systemic risk.
The concept of agentic AI itself demands careful definition for regulators accustomed to overseeing human actors and static software systems. A traditional algorithmic trading system executes pre-programmed instructions. An agentic AI, by contrast, pursues defined objectives through dynamic, context-sensitive decision chains — identifying opportunities, executing multi-step strategies, and revising its own approach based on real-time feedback. In financial markets, this distinction matters enormously. When multiple agentic systems interact simultaneously across interconnected markets, the potential for emergent, unpredictable behaviour — including cascading failures of the kind that produce flash crashes or liquidity crises — escalates in ways that conventional stress-testing models cannot fully anticipate.
The alarm being raised by central bankers reflects this systemic dimension above all else. Individual firms deploying agentic AI may manage their own model risk with considerable sophistication, but the aggregate behaviour of competing autonomous agents operating across the same market microstructure is a different problem entirely. This is the financial-stability version of what computer scientists call emergent complexity: individual agents following rational local rules that collectively produce irrational, destabilising system-level outcomes. Regulators at institutions including the Bank for International Settlements and the European Central Bank have been quietly developing analytical frameworks to model these interactions, but the field remains nascent relative to the speed of deployment in live markets.
Rathi's emphasis on new tools is therefore both a practical and philosophical proposition. On the practical side, regulators will need supervisory technology capable of monitoring agentic systems in near-real time — not quarterly filings or annual audits, but continuous surveillance of AI decision logs, objective functions, and market footprints. On the philosophical side, it requires regulators to reposition themselves as active participants in an evolving technological ecosystem rather than as external enforcers applying fixed standards to a static industry. This is a meaningful institutional shift for bodies whose authority has traditionally derived from their independence from market participants.
The collaborative model Rathi is advocating would likely involve structured information-sharing agreements between regulators and AI developers, regulatory sandboxes specifically designed for agentic systems, and joint stress-testing exercises in which firms and watchdogs simulate adversarial market scenarios together. Several major financial centres — including Singapore and the European Union — are already experimenting with analogous frameworks under their respective Monetary Authority of Singapore and European Banking Authority mandates, though none has yet produced a comprehensive agentic AI governance standard.
What This Means for Financial Institutions
For banks, asset managers, payment processors, and fintech firms already investing heavily in agentic AI capabilities, the regulatory direction of travel is becoming clearer even if the destination remains undefined. Institutions that wait for prescriptive rules before engaging with the oversight question are likely to find themselves on the wrong side of a collaborative relationship that regulators are explicitly trying to build now. The firms that participate constructively in regulatory dialogue — sharing data, stress-test results, and model documentation proactively — will almost certainly help shape the standards that eventually govern the entire industry. Conversely, those that treat agentic AI deployment as a purely internal risk management question, quarantined from regulatory engagement, may face far more disruptive intervention once formal frameworks crystallise. Rathi's message is clear: the window for collaborative norm-setting is open, and the regulators intend to use it.
Written by the editorial team — independent journalism powered by Codego Press.