Transparency, Talent, Trading Hours and the Future of AI Infrastructure
This week, AI in financial markets is one step closer to next week’s regulatory and operational tipping point. From the EU’s mandatory data disclosures to structural changes in talent development and market operating hours, here’s what I’ve learnt—and why I think it matters for the future of AI in trading.
- Get Ready for August 2: EU Mandates Public Disclosure of AI Training Data – Article 53(1)(d)
From August 2, all providers of general-purpose AI models governed by the EU AI Act will need to publicly disclose a detailed summary of the datasets used in training. This includes commercial and open-source models, and the format must follow the template provided by the EU AI Office.
Under Recital 107, companies must list the primary datasets and explain how the data was used, targeting claims many AI companies have relied on such as “we used publicly available data.”
Why it matters for trading: As AI becomes central to analytics, execution, and surveillance, the legal provenance of training data is now a data governance issue. These new rules will require firms (especially those integrating third-party models) to understand who trained what, using which data, and how it flows across the organisation.
- A Practical Example: AI Agents for Real-Time Financial News Summarisation
While the push for AI transparency ramps up, developers are already putting it into practice. Take this recent example from Hanane who built a self-improving financial news bot using OpenAI’s Agents SDK and an LLM-as-a-Judge loop to verify the final output.
Why it matters for trading: As AI transparency becomes non-negotiable, tools like this show how AI can deliver automation with control, creating systems that are explainable—enabling a shift from automation to domain-specific, context-aware intelligence, interpreting market data and financial disclosures into actionable ideas.
- Full EU AI Act Enforcement Begins August 2 – Governance, Risk, and Compliance
The training data rule is just one part of what becomes enforceable under the EU AI Act this coming week. The following is also in scope:
- Articles 28–39: Notification and documentation for authorities and notified bodies
- Articles 51–56: Rules for general-purpose models, including classification and systemic risk
- Articles 64–70: Governance frameworks at the national and EU levels
- Article 78: Confidentiality requirements
- Penalties, excluding fines for GPAI providers
EU AI Act Full Regulation (CELEX)
Why it matters for trading: the growing use of AI in portfolio construction, execution assessment and market oversight are now subject to binding regulatory obligations whether in-house or via third parties, making transparency, traceability, and risk frameworks no longer optional. Despite the perceived rollback by the US, firms may opt to raise internal governance to meet the Brussels Effect rather than risk the punitive fines which can reach €35 million or 7% of global annual turnover, whichever is higher.
- We need to talk about Devin – Goldman’s Rising Agents vs. the challenge of a lost generation
Goldman Sachs is piloting Devin, an autonomous AI software engineer developed by startup Cognition. Devin is already working alongside the bank’s approximately 12,000 human developers, with plans to scale from hundreds to potentially thousands of instances. According to CIO Marco Argenti, this agentic approach to software development is projected to deliver 3–4× productivity gains over previous AI coding tools—underscoring how seriously Goldman and the broader industry are embracing full-scale automation. Goldman isn’t alone. JPMorgan, Citi, and BofA are also rolling out AI assistants like Maestro, Client360 to streamline research, compliance, and internal workflows.
But this transformation raises a fundamental problem: if entry-level roles are being automated, who will develop the next generation of senior talent? Known as the AI-Becker problem, this challenge questions how firms will maintain a pipeline of capable supervisors, risk owners, and strategists if they no longer invest in training.
Why it matters for trading: as with any other industry role—talent pipelines don’t rebuild themselves. Just looking back to the perceived need to replace all junior traders with quants missed the mark, trading desks still need entry-level exposure to ensure oversight and governance of AI trading tools.
- AI and the Shift Toward 24/7 Trading
Following on from last week’s newsletter on LSEG’s move to 24/7 trading, SIX Group’s acquisition of Aquis targeting 20% of European equity trading signals a further push toward always-on infrastructure supported by AI-powered matching engines and smart routing.
Why it matters for trading: Around-the-clock trading introduces new liquidity patterns, pricing risks, and governance challenges. As AI systems take on more responsibility in routing, quoting, and execution, strong oversight frameworks will be essential to ensure fairness, efficiency, and stability—emphasizing that AI in trading is no longer experimental, it’s becoming structural.
Having written this blog for the last 6 months, it’s clear that the pace of change is accelerating. But with that comes the need for oversight, accountable design, and long-term thinking. The challenge isn’t just building the tech—it’s ensuring we have the talent, tools, and governance to keep up with it.
As always, thank you for reading! Let me know what you found helpful, what you disagreed with and what you’d like to see more of.
Many thanks,
Rebecca
