87% of Fed Market Contacts Cite AI Risk as Regulators Turn to Trading Data Quality
Within a few weeks in November 2026, three financial authorities published material on artificial intelligence in capital markets. The US Federal Reserve's Financial Stability Report identified AI as the risk most frequently cited by its market contacts. India's SEBI said it will soon issue guidelines on the responsible use of AI. Hong Kong's SFC released its first Large Firm AI Readiness Survey. Read alongside recent academic work on machine learning in markets, these developments point to a shared question: how reliable, traceable and well governed is the data AI models learn from, particularly records of trading behaviour?
| Date | Body | Development | Key figure |
|---|---|---|---|
| 13 November 2026 | US Federal Reserve | Semi-annual Financial Stability Report | 87% of 23 market contacts cited AI as a risk, up from 50% in spring |
| 14 November 2026 | SEBI (India) | Chairman announces forthcoming AI guidelines | Disclosure, accountability and a "human-in-the-loop" approach |
| Early November 2026 | SFC (Hong Kong) | First Large Firm AI Readiness Survey | 30% of firms use agentic AI; 25% are piloting it |
According to Bloomberg, the Federal Reserve on Friday, 13 November, flagged potential vulnerabilities stemming from artificial intelligence and elevated asset valuations in its semi-annual report on financial stability. The central bank said "Measures of equity valuations have remained high, partly driven by market enthusiasm for super intelligence," adding that this enthusiasm has been moderated by concerns about firms' capital expenditures and debt issuance plans.
The survey figures are the clearest signal. According to write-ups of the report, AI was the risk cited most often in the Fed's survey of market contacts, named by 87% of the 23 contacts, up from 50% in the spring survey. The report also details rapid growth in data center debt and generally reflects information available as of October 28, 2026. The Fed's concern centres on valuations and financing rather than model design. Even so, a move from half to nearly nine in ten contacts within six months shows how quickly AI has become a risk that market participants expect to manage.
In India, the securities regulator is moving from discussion towards rules. SEBI Chairman Tuhin Kanta Pandey announced on 14 November 2026 that SEBI will soon issue detailed guidelines for the responsible use of AI in capital markets. According to press coverage, the proposed framework will focus on disclosures, the accountability of regulated entities and a "human-in-the-loop" approach intended to keep critical AI decisions under human oversight. SEBI has not yet released a draft text, so these details rest on reporting of Pandey's remarks.
ETV Bharat reported that the guidelines will require companies using AI for stock market-related operations to provide full disclosure, and that a company will be held accountable for any AI-related irregularities. Pandey said the aim is to optimise benefits and minimise potential risks associated with integrating AI/ML-based applications in securities markets, thereby safeguarding investor protection, market integrity and financial stability. Under the proposed approach, no company will be permitted to rely solely on machines or AI for critical decision-making. SEBI is also applying AI to its own supervision and has developed supervisory tools including the "Sudarshan" tool.
Hong Kong's SFC published the findings of its first-ever Large Firm AI Readiness Survey, highlighting broad use of AI across business and control functions alongside challenges to scaling that use responsibly. Conducted between July and August 2026, the survey covered 20 large financial firms with 78 affiliated licensed corporations in Hong Kong: 12 large global firms, five large Chinese Mainland firms and three large online retail brokers.
| SFC survey indicator | Result |
|---|---|
| Firms already using agentic AI | 30% |
| Firms piloting agentic AI | 25% |
| Firms expecting AI use to accelerate within 12 months | 95% |
| Firms using or evaluating frontier models for cyber defence | 90% |
The reported use cases are relevant to market data. The SFC said the major use cases for agentic AI cover information retrieval, drafting, software development, operational workflows and some trading-related activities. Once agents touch trading-related processes, the quality and lineage of the data they consume moves from a research concern to a control concern.
The Dubai Financial Services Authority (DFSA) published related work in early November. Its report, Frontier and Agentic AI: Cyber Resilience in the Financial Sector, was prepared in collaboration with the Cambridge Digital Innovation and Regulation Initiative (C:>DIR). It arrives as brokers and trading technology providers begin giving AI assistants direct access to trading accounts. In financial services, the actions such agents could take include accessing customer information, changing system settings or submitting trading instructions.
Academic research explains why these regulatory moves lead back to data. In its editorial for a Journal of Financial Markets special issue, the journal states that as trading algorithms and machine learning models become primary actors in financial markets, the traditional frameworks of price discovery, risk management and investment strategy are being fundamentally affected. The papers in the issue identify novel predictors, from cryptocurrency order flow to granular environmental data, that often outperform traditional factors. The editors also point to the need for economic interpretation of machine-discovered patterns and for regulation of algorithms.
Order flow is a direct record of how market participants behave. A predictor built from it is only as useful as the underlying records are complete, correctly time-stamped and consistently defined. Missing or misaligned trades can produce patterns that appear meaningful but are artefacts of the data, which is precisely why the editors' call for economic interpretation matters.
A survey of large language models in financial prediction and trading, The New Quant, names the specific risks. It analyses challenges that matter in production, including temporal leakage, hallucination, data coverage and structure, deployment economics, interpretability, governance and safety. It calls for time-safe and economically meaningful evaluation that reports costs, latency and capacity, and closes with recommendations for standardising evaluation and building auditable pipelines.
| Data-quality issue | Effect on a model | Related regulatory theme |
|---|---|---|
| Temporal leakage (training data contains information not yet available at the time) | Backtests overstate live performance | Disclosure and accountability (SEBI) |
| Gaps in data coverage and structure | Weak generalisation across markets and periods | Responsible scaling (SFC) |
| Pipelines that cannot be audited | Outputs cannot be traced or explained | Human-in-the-loop oversight (SEBI) |
| Agents acting directly on accounts | Trading instructions executed without adequate checks | Cyber resilience (DFSA) |
The table above is an analytical mapping drawn from the sources cited, not a classification published by any regulator.
Earlier European evidence offers a baseline. In a February 2026 risk analysis, ESMA found a pronounced disparity in AI adoption across firm sizes. A national survey cited in the same report found 53% of respondents using AI or planning to use it within 12 months, almost identical to ESMA's own survey result.
None of these sources suggests that AI can reliably predict markets. The common thread is narrower: an AI system is only as reliable, auditable and accountable as the data it learns from. As regulators move towards disclosure requirements, human oversight and evidence of responsible scaling, well-curated and correctly time-stamped records of trading behaviour become part of what firms need to demonstrate sound governance of their models.
Key points
- The Fed's November 13, 2026 Financial Stability Report, as reported, shows 87% of 23 market contacts cited AI as a risk, up from 50% in spring.
- SEBI Chairman Tuhin Kanta Pandey said on 14 November 2026 that AI guidelines built on disclosure, accountability and human-in-the-loop oversight are coming.
- The SFC's first Large Firm AI Readiness Survey of 20 firms found 30% already use agentic AI, including in some trading-related activities.
- Journal of Financial Markets research finds novel predictors such as cryptocurrency order flow often outperform traditional factors, but calls for economic interpretation.
- An LLM quant survey flags temporal leakage, data coverage and structure, and auditability as production risks, linking model reliability directly to data quality.
- Bloomberg, AI Advances Raise Financial Market Vulnerabilities, Fed Report Finds, November 13, 2026
- CellCog, Fed Report: AI Is the Top-Cited Financial Stability Risk
- CASRAI, Fed report: AI valuations and data center debt
- Storyboard18, SEBI to issue AI guidelines for capital markets, prioritise investor protection
- ETV Bharat, Balancing Tech And Trust: Detailed View Of Sebi's Upcoming AI Guidelines for Capital Markets
- Indian Opinion, Sebi to issue AI guidelines for capital markets soon
- Securities and Futures Commission, SFC survey highlights wider AI adoption and challenges to responsible scaling
- CoinSpectator, Dubai Regulator Raises AI Trading Risks as Brokers Expand Automated Execution, November 4, 2026
- Journal of Financial Markets (ScienceDirect), The AI and machine learning revolution in financial markets
- arXiv, The New Quant: A Survey of Large Language Models in Financial Prediction and Trading
- ESMA, AI adoption and trends in securities markets: EU evidence, February 2026
