Dominic Jainy stands at the forefront of the modern digital revolution, bringing years of seasoned expertise in artificial intelligence, machine learning, and the decentralized architecture of blockchain. As an IT professional who has watched these technologies evolve from experimental theories to the backbone of global commerce, he possesses a rare perspective on how automated systems interact with legacy structures. His current focus lies in the high-stakes intersection of financial technology and regulatory oversight, where the rise of autonomous “agentic” AI is challenging the very foundations of how we maintain market stability. In this discussion, we explore the shifting landscape as major institutions like the Bank of England grapple with a world where software no longer waits for a human to press a button, but instead makes its own calls in the blink of an eye.
The conversation centers on the emerging regulatory gaps created by AI systems that operate with independent agency, moving beyond simple automation into complex decision-making. We delve into the critical risks of cyber resilience in a landscape where malicious actors and security teams use the same advanced tools, the systemic danger of correlated failures across shared digital infrastructures, and the potential for new market safeguards like “kill switches” to prevent AI-driven volatility.
Traditional financial regulations rely on human oversight, yet agentic AI often operates without direct instruction. How does this shift from human-led automation to truly autonomous systems change the way we must think about accountability in finance?
The fundamental shift we are witnessing is the departure from a world where we could always point to a person behind a specific trade or decision. In the current landscape, as Deputy Governor Sarah Breeden has highlighted, our existing regulatory frameworks were simply not designed for agents that act without a direct human prompt. It is no longer practical to expect a human to supervise every single action these systems take because the velocity and volume of those actions exceed our cognitive capacity. When a system can chain together sequences of tasks at extreme speed and scale, it stops being a tool and starts being a participant. This forces us to rethink accountability not as a chain of commands, but as a system of engineering and governance that must be baked into the AI’s objective functions from the very first line of code.
With recent reports indicating that 81% of financial services firms are adopting AI and 52% are already moving toward agentic systems, what do you see as the biggest operational hurdle for firms trying to integrate these autonomous agents safely?
The numbers from the 2026 Cambridge Centre for Alternative Finance report are staggering because they show that autonomy is no longer a niche experiment; it is becoming the industry standard. The biggest hurdle is that while 52% of firms are diving into agentic AI, most of that work is currently siloed in internal functions like software engineering and data visualization. Moving these systems into the “front office” of trading and payments is where the real friction begins, as firms must ensure these models don’t “drift” from their original purpose. There is a palpable tension in the industry as we try to build safeguards that are robust enough to stop a rogue model but flexible enough not to stifle the efficiency gains that 81% of the sector is chasing. We are essentially trying to build a high-speed engine and a new type of brake simultaneously.
The concept of cyber resilience has taken on a new dimension with agentic AI. How concerning is the “step change” in capability where AI can now chain together sequences of actions to find vulnerabilities?
This “step change” is perhaps the most visceral concern for anyone in IT security today because it levels the playing field between defenders and attackers in a terrifying way. When we talk about agentic AI being able to identify cyber vulnerabilities and autonomously chain attacks, we are describing a threat that can probe thousands of entry points in the time it takes a human to blink. The Bank of England is rightly worried that while these tools strengthen our defenses, they also give malicious actors a weapon that can create mass disruption across the entire financial system. It feels like a high-speed chess match played in the dark, where the risk isn’t just an outage at one bank, but a cascading failure across the shared digital infrastructure that supports our entire economy. We have to move past looking at individual firms and start stress-testing the resilience of the whole network.
There is a significant debate regarding the safety of open-source versus closed-source AI models. How much comfort should we take in the fact that open-source models currently trail the most advanced closed models by about four to eight months?
A four-to-eight-month lag provides almost no comfort in the grand scheme of financial stability, as it is a mere heartbeat in the timeline of regulatory development. This narrow window means that highly sophisticated, autonomous capabilities become available to anyone with an internet connection shortly after they are pioneered by the tech giants. If a malicious actor gets hold of a model that is nearly as powerful as the industry standard, they can launch attacks that several institutions might not be prepared for at the same time. The International Monetary Fund has warned that this creates a “correlated failure” risk, where a single targeted strike on shared cloud or payment networks could paralyze multiple sectors. We are dealing with a situation where the expiration date on our security advantages is less than a year, which is a frantic pace for any regulator to keep up with.
If autonomous systems begin to respond to market signals in similar ways, it could lead to extreme volatility. What is your take on the proposed use of “kill switches” and “circuit breakers” specifically designed for AI-driven markets?
The idea of a “kill switch” for a market sounds like something out of a sci-fi thriller, but it is becoming a necessary reality as we see models trained on similar data sets starting to move in lockstep. If dozens of autonomous agents all identify the same “opportunity” and react simultaneously, they can create a feedback loop that destroys value in seconds. The Bank of England is exploring whether we need market-wide guardrails to stop trading if these models contribute to severe disruption. It is a controversial move because it interrupts the “natural” flow of the market, but when the actors are non-human entities that might have drifted from public policy goals, a manual override is the only way to prevent a total meltdown. We are looking at a future where a central bank might have to “unplug” a market to save it.
The Financial Stability Board recently proposed 12 sound practices for AI adoption, emphasizing organization-wide governance. In your view, will these non-binding standards be enough to steer the industry toward responsible innovation?
These 12 practices are an excellent starting point because they force firms to define clear roles and responsibilities for AI, especially in critical functions, but non-binding standards always face the “race to the bottom” problem. When firms are under pressure to perform, there is a strong temptation to cut corners on governance in favor of speed and profit. However, the FSB’s focus on third-party risks and ICT resilience is crucial because no firm is an island in the modern fintech ecosystem. Even if these standards aren’t binding laws yet, they set the “tone from the top” that will eventually influence insurance premiums, audit requirements, and investor confidence. The real test will be whether firms treat these practices as a “check-the-box” exercise or as a genuine blueprint for the survival of their institution in an AI-dominated world.
What is your forecast for the future of central banking as agentic AI becomes more deeply embedded in our financial core?
I believe we are heading toward a revolutionary “failover” model where central banks will act more like the ultimate IT system administrators for the nation. We will likely see a shift toward the Bank of England’s suggestion of “recovery requirements” where one bank could potentially take over another bank’s basic functions during a massive AI-driven outage. This means that in the next five to ten years, the success of a central bank won’t just be measured by interest rates or inflation targets, but by their ability to rebuild compromised core systems in real-time. We are moving away from a world of isolated outages toward a landscape of mass disruption risks, and the authorities that succeed will be those that have the separate, hardened systems ready to take over the moment the “fog of the machine” becomes too thick to navigate.