Artificial intelligence is increasingly embedded in how banks onboard clients, operate and monitor transactions and manage financial crime risk. Its growing use raises an important question for banking litigation: where an AI system identifies a risk that a human decision-maker does not act upon, or conversely, unjustifiably fails to flag an obvious risk, how does that affect the scope and application of the Quincecare duty?

Recent commentary outside the legal sphere has framed AI as a “copilot” to human professionals, capable of analysing information with a level of precision that may exceed individual judgment. In financial services, this is no longer a theoretical proposition. Banks, particularly digital banks, are already relying on AI‑driven systems to detect unusual activity and recognise abuse and indicators of fraud. Against the backdrop of significant regulatory fines imposed on Monzo and Starling for weaknesses in financial crime controls, enhanced technological oversight is increasingly presented as a core component of effective risk management rather than a future innovation.

The Quincecare duty after Philipp

The legal framework remains anchored in the Supreme Court’s decision in Philipp v Barclays Bank UK plc [2023] UKSC 25, which is the definitive authority on the Quincecare duty in England and Wales. The court confirmed that a bank’s obligation to refrain from executing a payment arises only where the instruction is given without actual or apparent authority. The duty is not a free‑standing obligation to prevent fraud, but an application of the bank’s general duty to interpret, ascertain and act upon its customer’s instructions with reasonable skill and care.

Philipp, therefore, places firm limits on the scope of Quincecare claims. However, it does not freeze or fix how the standard of reasonable skill and care will be assessed in different situations, nor does it prescribe what a bank is taken to know in an evolving technological environment.

AI, reasonable skill and care, and being put on inquiry

Where banks deploy AI as part of their client take-on and account monitoring architecture, questions arise as to how that deployment informs the assessment of whether and when the bank was put on inquiry for Quincecare purposes. By way of example:

• If an AI system that is fully integrated into a bank’s processes generates an alert suggesting a lack of authority, the failure to act on that alert may carry legal significance. An ignored signal of that nature may operate as the modern equivalent of a traditional red flag.
• Furthermore, where effective account monitoring relies on the proper operation of an AI system, difficult questions will arise if the system itself is faulty or unfit for its intended purpose. Should a bank be held liable where it is established that relevant information (that would have put a reasonable bank on inquiry) was clearly captured within the system and was therefore technically “known” within the bank, but due to system failings, no appropriate alert was generated?

There is no suggestion that AI expands the Quincecare duty beyond the limits set by Philipp. Rather, it has the potential to recalibrate the point at which a bank is treated as having been put on inquiry. The relevant issue is not whether the bank could, in theory, have detected a problem using more advanced technology, but what the bank’s own systems actually identified or “knew,” and how that information was handled.

Evidential implications in litigation

A critical distinction must therefore be drawn between the theoretical capabilities of AI and its actual deployment within a bank’s control framework. The standard of reasonable skill and care is not and should not be elevated simply because more advanced technology exists in the market. However, once a bank chooses to integrate AI into its decision-making or monitoring processes, both the inputs and outputs of that system form part of the factual matrix against which the bank’s conduct will be assessed.

This has important evidential consequences in Quincecare litigation. Model outputs, alert logs, training parameters and escalation pathways may become central to disclosure and expert evidence. In appropriate cases, a claimant may not need to establish what a human operator noticed, but instead what the bank’s own systems identified and whether those signals were followed by reasonable process and inquiry.

Practical challenges

The difficulties of establishing liability, and in particular of demonstrating that a bank was put on inquiry, should not be underestimated. Claims of this nature increasingly turn on complex issues of system design, escalation logic and institutional knowledge, as well as the interaction between technology and established legal standards of care.

Specialist expertise is often required to navigate both the legal and technical dimensions of these disputes. With the ever-increasing rate of technological change, there comes a risk that “expertise” may quickly become outdated. The challenges associated with identifying an individual with the requisite understanding of current practices should not be underestimated.

How Stewarts can help

Stewarts is the UK’s largest disputes-only law firm, with more than 90 partners. As a litigation-only firm with no transactional departments, we are able to act against the largest banking institutions. Our team are widely recognised as a “go to” practice for banking and finance litigation as well as contentious insolvency disputes, and we are frequently instructed to investigate and advise on Quincecare claims. Our specialist team draws on expertise from senior litigators, former in-house banking lawyers and ex-investment bankers. Quincecare claims that Stewarts has acted on in recent years include: Arena Television Limited (in liquidation) v Bank of Scotland plc & Lloyds Bank plc; Kijani Resources Limited (in liquidation) and Ratio Limited (in liquidation) v Royal Bank of Scotland International Limited; and Stanford International Bank Ltd (in liquidation) v HSBC Bank PLC as well as advising on a number of further claims not in the public domain.