AI matters | Financial services round-up (May 2024)

If the mood music around AI developments was marked with musical terminology, then the annotation prestissimo – the fastest possible tempo – would not be far off the mark, evidenced by the launch of a new version of Open AI's GPT-4o (apparently it can laugh, chat and sing!) and news of the European roll-out of Anthropic's Claude AI assistant last week and, of course, with the staging this week of the AI Seoul Summit.

When it comes to regulatory announcements in the financial services industry, a sector where AI plays a fundamental role, while the metronome setting is a notch or two lower, there are nonetheless a number of interesting developments of note.

Monsters in the deep? | AI and financial stability

In a speech delivered, on 7 May 2024, Jonathan Hall, Financial Policy Committee External Member, considered whether the use of AI models created potential risks to financial stability. 

With specific reference to ‘deep learning’ (the engine behind AlphaGo and Large Language Models referred to above), Hall draw attention to two risks: 

• Deep trading agents might cause a less resilient financial market.
• There may be misalignment of the incentives of deep trading agents with those of regulators and the public good.

Hall emphasises the need for regulators, market participants and AI safety experts to collaborate to mitigate the harm of any future algorithms and draws attention to three areas of focus for future work:

• Training, testing, monitoring and control of deep trading algorithms.
• Alignment of algorithms with regulation.
• Stress testing - new stress scenarios are needed using adversarial techniques. 

While Hall's views are his own, it contains a number of useful pointers as to how the regulators are approaching this topic, as well as news that the Bank and FCA are carrying out a survey on machine learning survey this month.

DRCF's AI and Digital Hub pilot

Formed in 2020, the Digital Regulation Co-operation Forum (DRCF) brings together four UK regulators which have responsibilities for digital regulation: the Competition and Markets Authority, the Financial Conduct Authority (FCA), the Information Commissioner's Office and Ofcom. 

On 22 April 2024, the DRCF launched its AI and Digital Hub, a free service which will provide informal advice and support to innovators working on AI or digital products, such as which regulatory requirements apply to a product.  The initiative is in response to research commissioned by the DRCF that showed that innovators want streamlined regulatory support to assist them with releasing their AI and digital products to the market safely and quickly.

To be eligible for assistance from the Hub, the question being asked must fall within the remit of at least two of the regulators in the DRCF and the new product, service or business model must have an AI and/or digital focus, be innovative, and benefit consumers, businesses and/or the UK economy.

The DRCF will interpret "innovative” to cover both radical and incremental developments to a product, service or business mode.  The DRCF will publish case studies (appropriately anonymised and cleansed of confidential/sensitive information) based on queries addressed by the Hub to inform applicants and widen access. 

Regulators explain their ‘strategic approach’ to AI

And finally, back in February, we reported that the UK government had quizzed the financial regulators on their strategic approach to AI.  On 22 April 2024, the regulators published their responses:

• FCA response 
• Bank of England and PRA response

Both responses set out how the regulators' current approach is aligned with the government's voluntary guidance (which sets out five principles that should guide AI regulation) and detail the various AI and ML initiatives the regulators are planning for the year ahead. For its part, the Bank of England's response includes those areas where further clarification on its regulatory framework would be helpful.  These are: data management, model risk management, governance, and operational resilience and third-party risks.