By Dr Iain Brown, Global Head of AI & Data Science at SAS
Financial services is the sector moving fastest on agentic AI – intelligent systems that can independently plan, adapt and execute complex, multi-step tasks without constant human direction. Agents built this way are already live in fraud detection, compliance monitoring, and settlement reconciliation at some institutions. Adoption is still uneven though as most firms remain in pilot, rather than production.
According to UK Parliament data, around 75% of UK financial services firms currently use AI in a broader sense – typically LLMs. Uptake of autonomous systems specifically will increase at pace once firms have the right controls in place and the FCA’s forward-looking assessments already project that certain operational domains will reach a level where they outperform human reasoning.
The whole thing comes down to one question: who owns the risk? These agents don’t just process data anymore, they influence P&L directly, with authority over revenue and capital allocation that software has never had before.
Accountability cannot be delegated to autonomous systems. Financial leaders need to retain ownership of the decisions, outcomes, and risks their deployment creates. Kill switches, comprehensive monitoring, and strong observability are what make that possible, keeping humans firmly in control, and keeping trust intact across an increasingly autonomous financial ecosystem.
Governance often gets framed as a brake on adoption. In practice it’s the opposite: the firms moving fastest have the strongest controls, because trust is what gets a board to sign off pilots.
What gets overlooked is the back-office pressure building underneath. Faster front-end interactions just push the bottleneck downstream into settlement, reconciliation, and compliance checks that are still running at human speed. Even with layered governance frameworks in place, the risk of agent outages across the industry increases as a result.
The cost of downtime
Downtime carries an immediate financial and operational cost. On average, £11,000 is lost per minute by UK businesses – that’s why operational resilience in core banking matters so much. Downtime no longer means a system is briefly unavailable; it means entire operational chains, payment processing, trade settlement, fraud monitoring, customer servicing, can all be disrupted simultaneously.
This isn’t totally hypothetical either. In a 2025 Crypto trading competition that gave autonomous AI agents real trading capital and no human intervention, most lost money. One model dropped 63% of its capital, others around 30–56%. While it’s not exactly like a traditional banking operation, an autonomous agent making live trade and risk decisions, with far more complexity and far less predictability than that competition involved, represents an even bigger unknown and is a strong warning for firms to get it right.
When autonomous software built to handle thousands of tasks glitches, even briefly, the knock-on effect is significant. Even with more people on hand, the scale and speed of autonomous activity means a widespread failure could generate a backlog no team clears in time.
No autonomous system has yet caused a market-wide event on the scale of Knight Capital in 2012. But Knight Capital was a simple, rules-based algorithm operating within known parameters and it still took down a 17-year-old firm in 45 minutes. Today’s autonomous agents introduce more complexity and less predictability than that. It’s this shift in operational risk that demands a corresponding shift in corporate governance.
Restructuring corporate governance
Traditional corporate governance was built to audit simple, predictable algorithms. Agents that change their own path in real time need something else. That dynamic adaptability is exactly why new roles, specifically engineered to monitor autonomous software, are needed across the corporate landscape. Done well, this isn’t governance for its own sake, it’s what gives firms the confidence to deploy these systems at the pace the opportunity demands, rather than holding back out of caution.
One emerging concept is the Chief Agent Officer (CAO) – a role built specifically to give autonomous agents room to operate at scale, within defined business, risk, and compliance parameters. No major financial institution has created this exact title yet, but the groundwork is visible: in 2024, the US Office of Management and Budget required major federal agencies to appoint a Chief AI Officer, and firms including Lloyds, NatWest and Mastercard have since created their own AI leadership roles. As agent deployment scales past what those roles were designed to cover, a narrower, agent-specific mandate looks like the logical next step.
That oversight matters because autonomous systems can execute decisions at a speed and scale beyond traditional operational controls. An incorrectly configured agent, for instance, could apply pricing concessions or policy exceptions across thousands of customer accounts, creating financial consequences long before anyone catches it.
Preventing that requires organisations to build new operational capabilities alongside new technologies: training teams to audit autonomous decision paths, validate model inputs, and oversee increasingly complex interactions between AI systems.
That training matters because the autonomy itself is the risk. The software needs to be entirely programmable, so human operators can act the moment an agent does something outside the norm.
Take an autonomous corporate bond-trading agent tasked with optimising a portfolio: if a geopolitical event triggered a system crash, an unmonitored agent could continue executing acquisitions that expose a firm to millions of pounds in losses.
The organisations best positioned for what’s next won’t be the most cautious, or the fastest movers. They’ll be the ones that build governance and resilience from the outset. That combination is what lets well-designed agents deliver enduring value and adapt alongside technological and regulatory change. Capability without disciplined human oversight is a significant corporate risk in its own right; but oversight without the ambition to deploy at scale won’t achieve the value that this technology offers.
Get the ownership model right, and agentic AI becomes a genuine source of value with decisions the business can stand behind. That’s the redefinition financial services needs to make. Not who builds the agent but who owns what it does.