The Financial Conduct Authority (FCA) has expanded its artificial intelligence (AI) testing initiative, welcoming Barclays, Experian, and UBS among the latest participants. These firms join the second cohort of the regulator's sandbox, which provides a controlled environment for financial institutions to experiment with AI applications in real-world conditions while receiving regulatory support and oversight.
The initiative builds on the first cohort, which included major players such as Lloyds Banking Group, NatWest, and Monzo. The FCA's sandbox is designed to help companies that have advanced their AI development to test their innovations safely before full-scale deployment. Participants explore key questions around risk management, live monitoring, and the responsible deployment of AI for consumers and markets.
Focus areas and use cases
According to the FCA, the second group will test a range of customer-facing and business-to-business use cases. These include AI-enabled targeted support for investments, credit score insights for consumers, agentic payments, and money laundering detection. The regulator noted that banks are experimenting with diverse technologies, including agentic AI, small language models (SLMs), and emerging techniques such as neurosymbolic AI.
Jessica Rusu, chief data, information and intelligence officer at the FCA, stated: "We're continuing to collaborate with firms to support the safe and responsible development of AI in UK financial markets." She added that tailored support from the FCA and its technical partner Advai reflects the regulator's commitment to keeping pace with AI advancements and demonstrates how regulators and industry can work together to harness innovation responsibly.
The FCA plans to publish a report later this year highlighting both good and poor practices observed during the sandbox sessions. This report is expected to provide guidance for the broader financial sector on AI deployment.
Background on AI regulation in UK finance
The expansion of the AI sandbox comes amid growing scrutiny of UK financial regulators' approach to AI. Earlier this year, the Treasury Committee of the House of Commons criticised the FCA and the Bank of England for taking what it described as a "wait-and-see" approach to AI regulation. In a report, MPs warned that "the UK public and the country's finance system are exposed to potential serious harm because regulators in the financial sector are not doing enough."
The committee's chair, Meg Hillier, emphasised the urgency of proactive regulation, particularly after recent developments such as Anthropic's Project Mythos, which unearthed decades-old vulnerabilities in software systems. Hillier said: "Recent developments in the world of AI, such as Anthropic's Project Mythos, show us how fast this transformative technology is moving. It has never been more important that those responsible for maintaining the UK's financial stability take a proactive approach to understanding and mitigating the risks AI may pose to our financial system."
In response, Sarah Breeden, deputy governor for financial stability at the Bank of England, rejected the characterisation of a "wait-and-see" approach. She stated: "Far from taking a 'wait-and-see' approach, we have invested heavily in analysing the current and future risks posed by both the use of AI in financial services, and the broader investment in and adoption of AI across the wider economy." Breeden noted that the Bank shares the committee's view that AI has broad, complex, and likely long-term implications for how the UK financial system serves the real economy.
Last week, major UK banks entered discussions with regulators, finance authorities, and national security organisations following the release of Anthropic's latest AI model, Mythos. The model identified thousands of software vulnerabilities that had remained hidden for decades, raising concerns about systemic risks in financial infrastructure.
Industry perspectives and historical context
The FCA's sandbox initiative reflects a broader trend among financial regulators worldwide to create safe spaces for AI experimentation. Similar programs exist in the United States under the Office of the Comptroller of the Currency and in Singapore through the Monetary Authority of Singapore's regulatory sandbox. The UK's approach has been characterised as collaborative, aiming to balance innovation with consumer protection.
The financial services sector has been an early adopter of AI, using machine learning for fraud detection, credit scoring, algorithmic trading, and customer service chatbots. However, the increasing sophistication of AI, particularly generative and agentic models, introduces new risks such as bias, lack of explainability, and potential for market manipulation. Regulators are grappling with how to oversee these technologies without stifling innovation.
The FCA's sandbox allows firms to test AI applications in a controlled environment where the regulator can monitor outcomes and provide feedback. This iterative process helps identify best practices and potential pitfalls before widespread deployment. The inclusion of diverse use cases in the second cohort, ranging from investment advice to anti-money laundering, demonstrates the breadth of AI applications in finance.
Challenges and criticisms
Despite the proactive steps, critics argue that the sandbox approach is too limited. The Treasury Committee's report highlighted that only a small number of firms can participate, leaving many smaller players without access to regulatory guidance. Additionally, the voluntary nature of the sandbox means that firms not participating may deploy AI with less oversight.
The committee also pointed out that the FCA and Bank of England lack formal powers to enforce AI-specific rules, relying instead on existing regulations that may not adequately address novel AI risks. For example, current rules around algorithmic trading were designed before the emergence of large language models and agentic systems that can act autonomously.
In response to these criticisms, the FCA has committed to publishing a report on good and poor practices from the sandbox, which could inform future regulatory frameworks. The regulator has also invested in internal AI expertise, including hiring data scientists and partnering with academic institutions.
Technological depth: agentic AI and neurosymbolic AI
The FCA noted that participants in the second cohort are experimenting with agentic AI, which refers to systems that can autonomously pursue goals and make decisions without human intervention. In financial services, agentic AI could be used for automated trading, portfolio management, and fraud response. However, these systems raise concerns about accountability and control, especially if they act in unexpected ways.
Small language models (SLMs) are another area of exploration. Unlike large language models (LLMs) such as GPT-4, SLMs are designed to be more efficient and specialised for specific tasks, such as analysing financial documents or generating compliance reports. They require less computational power and can be deployed on local infrastructure, reducing latency and privacy risks.
Neurosymbolic AI combines neural networks with symbolic reasoning, aiming to achieve both pattern recognition and logical inference. This hybrid approach could improve the interpretability of AI decisions, a key requirement for regulated industries like finance. For instance, a neurosymbolic system could explain why it denied a loan application by referencing specific rules and data points.
The FCA's report later this year will likely provide insights into how these technologies perform in real-world conditions and what safeguards are needed.
Global context and future outlook
The UK is not alone in grappling with AI regulation in finance. The European Union's AI Act, which came into force in 2024, classifies AI systems by risk level and imposes strict requirements on high-risk applications, including those in credit scoring and insurance. The US has taken a sectoral approach, with agencies like the Consumer Financial Protection Bureau and the Securities and Exchange Commission issuing guidance rather than comprehensive rules.
The FCA's sandbox model is seen as a middle ground, allowing experimentation while gathering data to inform future regulation. The inclusion of firms like Barclays, Experian, and UBS signals that large, systemically important institutions are taking AI seriously and seeking regulatory clarity before full deployment.
As AI continues to evolve, the debate over regulation will intensify. The Anthropic Mythos incident underscores the need for vigilance, as even advanced AI can uncover risks that have lain dormant for years. Financial regulators must balance the benefits of AI—such as increased efficiency, better customer service, and enhanced security—against the potential for systemic disruption.
The FCA's initiative, while relatively narrow in scope, represents a step toward evidence-based policymaking. By working closely with industry, the regulator hopes to develop a framework that supports innovation while protecting consumers and maintaining market integrity. The coming months will be critical as the second cohort completes its testing and the FCA publishes its findings.
Source: ComputerWeekly.com News