What I learnt from my conversations last week on the latest on AI in trading
This week’s musings cover:
latest from the Regulators
Improvements to data challenges
Interoperability and the rise of MCP
Vibe coding and
AI Model Verification
- Regulatory Focus Intensifies
Both ESMA and the SEC are stepping up their scrutiny of AI in financial markets this week.
ESMA have issued the latest TVP warning of vulnerabilities tied to AI adoption, such as data quality issues, third-party dependencies, and model governance risks (https://www.esma.europa.eu/). For those who are interested, there is an ESMA webinar on their latest findings – you can sign up here: https://www.esma.europa.eu/press-news/hearings/webinar-artificial-intelligence-eu-investment-funds-adoption-strategies-and
Meanwhile, the SEC is also hosting an AI roundtable on March 27, 2025 looking at AI’s evolving role in investment strategies, governance risks, and potential systemic vulnerabilities. The event is open to the public both in-person (registration required) and online without registration – https://www.sec.gov/newsroom/meetings-events/sec-roundtable-artificial-intelligence-financial-industry
- Data Remains the Foundation
AI’s effectiveness continues to hinge on high-quality data. The FIX AI Working Group is building a #FIX Tag dictionary to support this (for more information please message me direct). Meanwhile, #Microsoft’s partnership with #Inait is moving the data debate on by aiming to develop AI models inspired by mammalian brain reasoning, focusing on real-world learning rather than pure data correlation (https://on.ft.com/4ivnCbV) . This could drive improvements in trading algorithms and risk management tools.
- Another key area of concern has been interoperability. The MCP (Model Communication Protocol) is emerging as a potential advancement in addressing this challenge. MCP functions as a universal data connector for AI models, streamlining access to data across APIs, databases, and files — much like Wi-Fi enables seamless connectivity between devices. By simplifying data integration, MCP allows AI systems to efficiently access real-time data when supported by the source – read more here (https://medium.com/data-and-beyond/the-model-context-protocol-mcp-the-ultimate-guide-c40539e2a8e7) Similar to FIX and FDC3 the core theme is to reduce integration complexities, but also to enhance model reliability through consistent and verified data access.
- The Rise of “Vibe Coding” and Its Risks
The viral trend of “vibe coding,” where developers use AI to write code illustrates how future AI interactions may become more intuitive and results-driven, enabling faster iteration and feature development. However, some are highlighting additional risks. Recent linkedin articles highlight a lack of source control, resulting in critical failures, while others had no security, making them vulnerable to malicious attacks (read more here – https://www.linkedin.com/posts/uatuko_theres-a-lot-of-hype-about-vibe-coding-activity-7308073214079893505-tHey/). Services dedicated to fixing these flawed apps are emerging – so one to watch how this develops.
- AI Model Verification Gains Urgency
The increasing complexity of agentic AI is strengthening the need for robust verification methods. One solution is multi-agent systems, where multiple AI models collaborate to validate each other’s outputs, improving accuracy and reducing the risk of flawed logic (https://www.linkedin.com/feed/update/urn:li:activity:7308859083590127617/). The second approach gaining traction is neuro-symbolic AI, which combines deep learning’s pattern recognition with rule-based logic to improve decision-making. To ensure these systems behave as expected, developers often rely on integration-test-style “evals”, which test overall performance in controlled environments – again one to watch (https://www.linkedin.com/posts/larrytheliquid_the-increased-adoption-of-agentic-or-dynamic-activity-7305583085843021824-ke9M/?utm_source=share&utm_medium=member_ios&rcm=ACoAAARqE5IBeeTUq6qzU4Eqccq_UO6-OMeR6Ao).
As always, your feedback is always welcome — let me know what you thought of this weeks’ content and what topics would you like to see covered next?
Many thanks
Rebecca
