What is AI in trading?

AI is used in trading in each phase of a trade: the analysis before it, the decision, the execution and the processing after it. ESMA describes the role of AI in trading in its article on AI in EU securities markets (ESMA, February 1, 2023).

These AI trading examples come from brokers and buy-side firms. ESMA also cites an industry report: most algorithmic trading of banks and large non-bank market makers "is still built around relatively transparent rules-based models".

AI trading works under MiFID II as algorithmic trading whenever the model sets the parameters of an order. Article 4(1)(39) defines algorithmic trading as trading "where a computer algorithm automatically determines individual parameters of orders such as whether to initiate the order, the timing, price or quantity of the order" with limited or no human intervention. A system that only routes orders to a trading venue is excluded.

Article 17 then sets the duties of an investment firm that trades this way:

• systems and risk controls with trading thresholds and limits that prevent erroneous orders and a disorderly market
• business continuity arrangements and fully tested, monitored systems
• a notice to the competent authority of its home member state and to the trading venue
• for high-frequency trading, time-sequenced records of all orders, including cancellations

High-frequency trading is the variant with co-location or other low-latency access, no human intervention for single orders and high message rates (Article 4(1)(40)). The definition names no method, so an AI model and a fixed rule set fall under the same article.

AI quantitative trading is still rare in the EU compared with other uses of AI. ESMA and 15 national authorities surveyed financial market participants between June and September 2025; 728 entities answered and 395 firms reported 847 AI use cases.

The most common use was drafting and summarizing information, with 239 use cases or 29% of the total. Algorithmic trading had 10 use cases, high-frequency trading 3 and robo-advising 2. ESMA writes that such core investment uses "remain relatively rare, with most still in development or experimental stages" (ESMA, February 20, 2026).

One risk of AI trading that ESMA names is that many firms run similar models from the same few providers and then trade in the same direction. ESMA cites studies that such concentration may cause systemic risk in algorithmic trading through herding, similar investment strategies and chain reactions that increase volatility during shocks. It adds that there is no concrete evidence yet that AI drives this process.

In its statement of May 30, 2024, ESMA lists further risks for investment firms: algorithmic biases, data quality, opaque decisions by staff and overreliance on AI by firms and clients. On February 26, 2026 ESMA issued a supervisory briefing on algorithmic trading for national authorities. It covers pre-trade controls, governance, testing and outsourcing, and adds a section on the use of AI.

Trading algorithms are not on the high-risk list of the EU AI Act. Point 5 of Annex III covers credit scoring and the pricing of life and health insurance, among other uses.

For algorithmic trading in Germany, section 80(2) of the Wertpapierhandelsgesetz (WpHG) transposes Article 17 of MiFID II. A firm that trades algorithmically needs systems that are resilient under Chapter II of DORA and subject to trading thresholds and limits. BaFin supervises.

Deutsche Börse runs the primary back-ends of its trading venues Xetra and Eurex in the Equinix FR2 data center in Frankfurt. Trading firms can place their own servers in the same building through Deutsche Börse's co-location service, the low-latency access that Article 4(1)(40) of MiFID II names for high-frequency trading (Deutsche Börse).

In Austria, section 27 of the Wertpapieraufsichtsgesetz 2018 (WAG 2018) sets out the same duties. A firm that trades algorithmically must notify the FMA and the authority of the trading venue. The FMA can ask for a description of its algorithmic trading strategies.

Switzerland is outside the EU, so MiFID II does not apply there. Article 31 of the Financial Market Infrastructure Ordinance (FinMIO) requires trading venues to identify orders generated by algorithmic trading. Participants must flag such orders and have risk controls that keep their systems within trading thresholds. FINMA supervises. This page gives no legal advice.