Technology and Innovation Community

Exec Summary

Although the financial industry has been an early adopter of statistics and machine learning, the use of more advanced artificial intelligence (AI) in financial markets is still at an early stage. However, improvements in computing power and easier access to AI tools are accelerating adoption. Firms and regulators must therefore balance the potential benefits of AI with the risks of introducing it into existing systems.

This review looks at how AI is currently being used, the risks involved, and whether existing risk-control frameworks are adequate, with the goal of informing the discussion around AI in financial markets. While AI is increasingly used in back-office and support functions, this paper focuses specifically on trading applications. Based on discussions with industry practitioners, several key observations emerge:

1. Autonomy: AI used in trading is typically part of automated, scalable, data-driven systems, but these systems are not fully autonomous. Human supervision and the ability to intervene remain standard practice.

2. Use cases: AI is most often applied to specific components within larger trading systems—for example, tools that analyze liquidity, recommend trading venues, support pricing decisions, or generate trading metrics.

3. Model risk: AI makes it possible to address more complex tasks and objectives. The main risks usually stem from the scope and complexity of these tasks, rather than from the AI methods themselves.

4. Monitoring outputs: As AI models become more complex and harder to interpret, fully explaining their internal decision processes may not always be practical. Instead, greater emphasis should be placed on monitoring their outputs and applying independent controls proportional to the risks of those outputs.

5. Control frameworks: Many of the risks associated with AI are already addressed by existing frameworks, such as model risk management and real-time controls for algorithmic trading. However, new or large-scale AI applications may evolve faster than these safeguards, making regular updates necessary.

6. Human accountability: Clear responsibility must remain with human developers, traders, and managers for the actions of AI systems, consistent with the accountability standards that apply to traditional and electronic trading.

7. Future developments: Over the longer term, trading systems may adopt more advanced techniques, such as generative AI or even more autonomous forms of AI, potentially reducing the need for human oversight. However, highly autonomous systems and large-scale AI-to-AI interactions are not yet a reality in financial markets.