By Noyan Nihat, Co-CEO and Executive Director of Cardaq, a UK-based, FCA-authorised electronic money institution and principal member of Visa, Mastercard and China UnionPay, and Deputy Chair of The Payments Association Advisory Board.
A recent FCA report found that at one wholesale bank, inaccurate code recommendations produced by AI had been through human review without anyone spotting the errors, and suffered a loss of service as a result [1]. In addition, it highlighted that staff were using unapproved AI tools, resulting in data leaks of sensitive information. Firms have since introduced a host of new safety measures, including mandatory staff training and processes to declare and sign off on AI tools.
Whilst these AI findings only play a small part in the overall report, it’s an important finding and a cautionary tale for payment firms in a rapidly advancing age of AI technology.
The internal audit findings say more than the headline incidents
The important takeaway for any payments board is the weaknesses uncovered by the banks’ own internal audit team. It pointed to committees not tracking conditional approvals and also mistakenly recording them as full approvals. Risk ratings were also applied inconsistently, AI inventories were incomplete, and kill switches had not been properly tested, with no documented list of who was authorised to operate them.
The failings uncovered do not need an AI specialist to resolve them. They are ordinary governance failures of which every regulated firm already knows how to look out for. The problem is that the technology is currently moving faster than the systems and processes around them.
Most firms have the policy, but fewer have the follow-through
The Bank of England and the FCA published their joint AI survey a week before the FCA wholesale banks standalone report, surveying 196 firms [2]. 86% of financial services firms now use AI and, positively, 94% have a firm-wide AI policy in place. The most common control in 95% of cases is to have a human review and approve the higher-risk cases. Where this encounters problems is that ongoing monitoring, leakage testing, and periodic recertification were only in place for around half, and only 64% named their board as accountable. Lawyers at Lewis Silkin drew the same conclusion in their review of the survey, noting that controls are much stronger before deployment than after it [3].
Why this matters more for payments firms
I’ve spent the past decade building electronic money institutions, and my concern is that these weaknesses are likely to become more common and harder to absorb in financial services firms like ours compared to the wholesale banks that the FCA examined. Smaller payments firms can adopt new tools quickly and typically have fewer people in compliance and internal audit teams. Hence, a staff member pasting customer data into an unapproved chatbot is a serious problem at a large bank, but at a small EMI, the same problem can go unnoticed for a much longer period of time.
A recent report by UK Finance showed that £1.28 billion was lost to payment fraud last year, and APP fraud increased by 19% to £576.4 million, with two-thirds of cases originating online [4]. Firms were required to pay out £354.3 million in losses as a result. I believe now is the time that the industry needs to focus on using AI to fight AI-enabled fraud and utilise agentic systems to detect and block suspicious payments in real time. But it’s vital that fintech innovation does not outpace security measures.
Autonomy needs an off switch
There is a tension here that firms of all sizes need to address. Real-time fraud prevention only works if AI can act without waiting for a person to approve each decision that it makes. The joint survey findings suggest that the industry is still understandably cautious about this, with only 2% giving AI agents that kind of broad autonomy [2]. The FCA’s findings show that for adoption to increase, key measures need to be in place, including a tested kill switch, a named list of authorised users to operate it and an accurate inventory of what is running. An autonomous fraud system with these controls is far easier to defend to a regulator than an AI-generated code change that passed a review that wasn’t scrutinised accordingly.
The regulator has been clear about where responsibility sits moving forward. Speaking at Bloomberg’s UK Policy Series on 19 October, Jessica Rusu, the FCA’s chief data, information and intelligence officer, said the FCA will not write new AI rules but expects senior managers to be personally accountable for AI in their areas [5]. She also said the existing duty to report major incidents should extend to “significant AI failures and near-misses” [6]. Two days later, Emma Banymandhub, CEO of The Payments Association, warned that governance “cannot be something firms retrofit once an AI system is already live” [7].
Guidance for payment boards
- Check that every approval of an AI tool or AI-generated change is recorded accurately, including any conditions attached, and that those conditions are being met.
- Make unapproved AI use a standing board item that is reported alongside the model risk register so it does not take a leak to bring it to light.
- Run a disclosure amnesty before an incident forces one and repeat it at regular intervals.
- Test the kill switches on any AI that can act on its own and write down who is allowed to use them.
Scale in payments has never only been about processing more transactions. As it depends on the operational and regulatory foundations underneath, AI is now part of those foundations whether firms have planned for it or not. The firms that fix these basics now will be in a much stronger position when the FCA turns its attention to a framework for agentic AI.
Sources
- FCA, Wholesale banks supervision – Insights 3 (Autumn 2026), 2 November 2026.
- Bank of England and FCA, Artificial intelligence in UK financial services – 2026, 26 October 2026.
- Lewis Silkin, Key lessons from the Bank of England and FCA’s 2026 survey, 30 October 2026.
- UK Finance, Annual Fraud Report 2026, June 2026 (figures for calendar year 2025; reported by Payment Expert).
- Law News, FCA sets no new AI rules but tells senior managers they will be held to account, October 2026.
- The Register, UK finance watchdog wants firms to report AI failures and close calls, 20 October 2026.
- Insurance Edge, Quick comments from Payments Association on FCA’s AI regulation ambitions, 21 October 2026.


