Copy article

The use of AI in the financial system

ended 07. April 2026

In a report published last week (points 50-55), the Bank of England's Financial Policy Committee (FPC) said: “At present, there was little evidence that the financial system had adopted more advanced forms of AI in a manner that would present systemic risk. For example, there was little evidence that financial firms were using advanced forms of AI to make core financial decisions, such as credit and insurance underwriting, or in core trading and investment activities. The FPC noted that one key driver of this was firms’ assessment that the lack of interpretability and predictability in advanced AI systems meant the potential risks from deploying them in more material use cases exceeded the potential gains. However, and noting the considerable growth in AI capabilities in recent years, the FPC also judged that risks appeared likely to increase, amid growing intent among financial firms to expand their deployment of advanced AI. At some point, adoption could potentially accelerate significantly as the technology improved, shifting firms’ assessment of the associated costs and benefits." AI experts: do you agree with this assessment that there is, as yet, no systemic risk — or is the reality different?  And do you agree that risks, as the FPC concludes, are likely to increase? If so, how can these risks be managed?

3 responses from the Newspage community

Copy all

Star Quote
Copy

Editor
Katrina YoungVerified
Futurist | Chief Technology Officer CTO/CDO Ethical AI & Digital Transformation Strategist at KYC Digital
The FPC assessment focuses on where AI is not yet embedded, rather than where risk is already forming. Systemic risk does not begin at full automation. It builds as influence scales.

AI is already shaping financial decisions through internal tools, vendor systems and model-driven analysis. These systems may sit behind human oversight, but they are not neutral. They guide outcomes at scale.

The greater concern is convergence. Financial systems become fragile when many actors behave in similar ways at the same time. As firms adopt comparable models, trained on overlapping data, decisions begin to align. That is how isolated errors can spread quickly.

There is also an operational reality often overlooked. Oversight becomes more demanding as systems grow in complexity. Expecting consistent human challenge under pressure is not always realistic.

Risk is therefore less about whether AI is used, and more about how quickly its use scales across firms. The priority now is governance.
Copy

Colette MasonVerified
AI Ethics Consultant at Clever Clogs AI
The FPC is right that banks haven't handed core decisions to AI. But framing systemic risk as something that arrives when they do misses what's already building. Every AI deployment needs human oversight, and oversight has a cognitive cost nobody is accounting for. The more you scale AI across a firm, the more policing needed. People chopping between complex, high-stakes scenarios rather than steadily progressing one problem fully compounds. that mental load.

Compare a 1960s Mini with a 2020s model. Same name, completely different machine. Dozens of systems hidden behind fascia plates you can't see through. When something goes wrong, even the mechanic needs specialist diagnostics. Financial AI is heading the same way. The more moving parts behind black box interfaces, fewer people who understand what the tech is doing.

When foolhardy competitive pressure overrides AI system safeguards, everyone loses.
Copy

Rohit Parmar-MistryVerified
Founder at Pattrn Data
I broadly agree with the Bank of England's read for now: advanced AI is not yet embedded deeply enough in core underwriting, trading or insurance decisions to create an immediate systemic threat. Most financial firms still do not trust these systems with the decisions that can break balance sheets, and frankly that caution is sensible.

Where I would be more sceptical is the pace of change. Adoption rarely moves in a straight line. It stays cautious for years, then jumps once tooling improves, vendors package risk away and executives convince themselves the governance can catch up later. That is when trouble starts.

The main risks are concentration, opaque decisioning, model correlation and over-reliance on systems nobody can properly explain under stress. Managing that means human accountability, clear audit trails, scenario testing, limits on autonomous use in material decisions, and regulators asking one blunt question early: who can actually explain why the model did what it did?