Banks must prepare for AI failures before customers pay the price, UK watchdog warns

Britain’s financial regulator is warning financial institutions that increasingly autonomous AI systems could cause widespread disruption, expose customers to cyber threats, and make accountability harder to establish. It wants firms to report serious AI failures and near-misses before they turn into bigger problems.

Artificial intelligence is rapidly becoming part of everyday banking, from customer service chatbots to fraud detection and automated financial decisions. But what happens when an AI system makes a costly mistake, exposes sensitive information, or nearly triggers a disaster?

The UK's Financial Conduct Authority (FCA) wants financial firms to prepare for precisely those scenarios — and report significant AI-related failures, including incidents that were narrowly avoided.

Key points

• The FCA wants financial institutions to report significant AI failures and near-misses, not just incidents that cause actual damage
• 86% of financial firms now use AI, the regulator says, citing an upcoming joint survey with the Bank of England
• Autonomous AI agents could introduce new cybersecurity risks as they gain the ability to manage payments, investments and other transactions
• Firms are being urged to stress-test AI systems, maintain human oversight and prepare fallback procedures
• Consumers are increasingly using general-purpose AI chatbots for financial advice, often without understanding the limits of their protections

AI failures could become a regulatory priority

Jessica Rusu, the FCA's chief data, information and intelligence officer, warned that the financial sector's growing dependence on AI creates risks that existing safeguards must address.

"Poorly governed AI can create consumer harm, cyber vulnerabilities and operational failures, potentially at much greater speed and scale," Rusu said at Bloomberg's UK Policy Series on Oct. 19.

The regulator already requires firms to report major operational incidents. Rusu argued that this obligation must also cover significant AI failures and near-misses.

A near-miss might involve an AI system attempting an unauthorized transaction that is blocked by a security control, or a chatbot generating dangerously incorrect financial information that’s spotted before reaching customers.

Neither scenario necessarily inflicts immediate harm, but both could reveal weaknesses that might cause serious problems later.

The idea is similar to safety reporting in aviation, where investigating narrowly avoided accidents can help prevent future disasters.

The FCA's announcement signals a supervisory expectation, rather than a new, finalized AI-specific reporting regime.

When AI starts making financial decisions

The FCA is particularly concerned about the emergence of agentic AI — systems capable of taking actions on behalf of users rather than simply answering questions.

Unlike conventional chatbots, AI agents could compare financial products, initiate payments, manage investments and interact with other automated systems.

Such capabilities promise greater convenience but also introduce new security challenges.

An AI assistant that merely provides inaccurate information can mislead a customer. An autonomous agent acting on that information could potentially execute an unwanted transaction before anyone notices.

Compromised AI agents, manipulated instructions, excessive permissions and weak authentication could also create opportunities for cybercriminals.

For example, an attacker might try to manipulate an AI-powered financial assistant into changing payment details or disclosing sensitive customer information.

The regulator wants the industry to develop clearer safeguards around agent identity, delegated authority, accountability and transaction execution.

In other words, financial institutions need to establish not only what an AI system can do, but who is responsible when something goes wrong.

Banks are already making mistakes with AI

Alongside Rusu's speech, the FCA published examples of good and poor AI practices observed during its engagement with financial firms.

Some findings suggest that institutions are adopting AI faster than they’re developing safeguards to manage it.

The regulator identified firms that deployed AI tools with insufficient testing, introduced governance procedures only after deployment, or failed to maintain adequate records of automated decisions.

It also warned against treating human oversight as a formality.

In one example, reviewers increasingly accepted AI-generated recommendations without meaningful scrutiny, even when there were signs of errors.

The FCA highlighted better practices, including limiting chatbots to approved information sources, maintaining audit trails, and allowing customers to escalate problems to human advisers.

A system that works correctly during testing may behave differently when exposed to unusual circumstances, changing models or large numbers of simultaneous transactions.

Rusu urged financial firms to stress-test AI systems, understand their dependence on external technology suppliers, and maintain fallback options when automated services fail.

Consumers may not realize AI financial advice comes without protection

The regulator also raised concerns about consumers turning to general-purpose AI chatbots for financial guidance.

According to figures cited by the FCA, four in five less-experienced investors aged 18 to 40 have used AI to help with investing.

Yet only 40% of consumers recognize that acting on advice from a general-purpose AI tool doesn’t result in the same regulatory recourse as receiving advice from an authorized financial adviser.

The FCA is calling on AI providers to make these limitations clearer through consistent, proportionate warnings.

The watchdog wants consumers to understand when they are dealing with an authorized financial institution and when they are simply receiving information generated by an AI model.

How to protect your finances when using AI for financial decisions

AI can be useful for explaining financial concepts, comparing terminology or helping consumers formulate questions. However, it should not be treated as an unquestionable financial authority.

A few precautions can help reduce the risks:

• Verify financial recommendations independently. Check investment claims, fees, risks and product details against official sources before making decisions
• Check who you're dealing with. In the UK, use the FCA Firm Checker to establish whether a financial services provider is authorized
• Be cautious about granting AI agents financial access. Understand what an assistant can see, what transactions it can initiate, and whether it requires your approval
• Avoid sharing sensitive financial information. Don't paste banking passwords, authentication codes or unnecessary personal details into general-purpose chatbots
• Watch for AI-powered impersonation scams. Criminals can use convincing AI-generated messages, voices and websites to impersonate banks, investment advisers and support representatives
• Secure your accounts and devices. Enable multi-factor authentication, keep software updated, and use reputable security software to help detect phishing, malicious links and other threats

If an AI-powered banking service produces an unexpected transaction or suspicious recommendation, stop interacting with it and contact the financial institution through an official channel.

Bottom line

The FCA isn't calling for banks to abandon artificial intelligence. It wants them to recognize that the technology introduces risks that can spread faster than conventional operational failures.

As AI moves from answering questions to making decisions and carrying out transactions, the consequences of a mistake become harder to contain.

Reporting close calls, maintaining meaningful human oversight and testing systems against unexpected behavior could help financial institutions catch dangerous weaknesses before customers pay the price.

On topic:

UK warns powerful AI models could pose security risks before they're even released

AI agents want to shop for you; banks warn of scams and privacy risks

Britain’s big push to make the internet safer