An ECB discussion paper asks what AI does to price discovery
Serguey Asael Shinder2 min readThe ECB published Discussion Paper No 32, "Artificial intelligence and financial markets", in September 2026 — 86 pages by Álvaro Cartea, Jean-Edouard Colliard, Thierry Foucault, Peter Hoffmann, Rob Graumans and Jean-David Sigaux.
The paper's premise is stated in one sentence: AI is transforming financial markets by enabling investors and intermediaries to extract more information from increasingly abundant data, through advances in algorithms and computing power. From there it works through three channels — information production and decision-making, labour demand, and financial intermediation — with particular emphasis on securities markets, and then surveys the implications for market efficiency, competition, informational frictions and financial stability.
The non-technical summary adds the adoption picture: intermediaries are investing heavily in data, computing infrastructure, algorithms and specialised staff; survey evidence shows adoption is widespread and accelerating, especially among large firms and those active in securities markets; and finance jobs are among those most exposed to AI-driven automation.
The authors give the reason that is easy to skip past: information is at the heart of almost every financial activity. Intermediaries collect and process information to assess investments, manage risk and allocate capital — so a technology that changes the cost of processing information is not a tool the industry adopts, it is a change to the industry's input.
What it means
The interesting question in the paper is not whether AI makes markets more efficient, it is whether it makes them more concentrated. Both follow from the same mechanism. If extracting a signal from data gets cheaper, prices should incorporate information faster — but the fixed cost of the data, the compute and the staff is what the summary describes firms investing in heavily, and fixed costs favour size. "Especially among large firms" is the sentence that carries the tension.
The labour finding is the one most likely to be quoted out of context. "Most exposed to automation" is a statement about task composition, not about headcount, and a discussion paper surveying a literature is not forecasting job losses.
📌 A discussion paper represents its authors, not the ECB, and the series says so on every cover. This is a survey of what is known and argued, published to be argued with — the useful thing about it is the map, not a verdict.