Competition authorities' use of artificial intelligence in their casework has been reported for some time. But how do they use AI? For which use cases? And what does it mean for businesses? A new cross-agency report by Stanford's Computational Antitrust initiative provides an answer: 30 competition authorities describe, in their own words, which tools they use and what those tools have enabled them to do. This post gives an overview of the use cases and of what is happening in Austria and at the European Commission in particular.
The answer in short: the use of these tools now ranges from detecting cartels through the review of seized material and the pursuit of resale price maintenance all the way into merger control. More and more competition authorities are using AI for more and more purposes, and they exchange experience on which applications work.
Antitrust
Detection in publicly available data. Authorities continuously review data that is public – producer price indices, prices from online shops and comparison portals, fuel prices, official statistics – and now increasingly have statistical models search it for patterns that are – in the authorities' view – hard to explain by competition. These include, for example, price movements that cannot be explained by costs, exchange rates or demand, or an unusually narrow price dispersion combined with persistently high margins. Large volumes of data are searched by AI for anomalies, those anomalies are flagged and case handlers then review them manually. The Brazilian authority monitors local fuel markets in this way, which in 2025 led to more than twenty search warrants at filling stations.
Review of seized material. Dawn raids produce e-mails, chat histories and mobile device data in volumes that could previously be reviewed only through keyword searches and with corresponding manpower. New tools organise this material by communication relationships, open up foreign-language documents and flag those that match known patterns of conduct – price fixing, market sharing or exchanges of information. The Lithuanian authority has tested such a system: in two cases with fragmented evidence it found, by its own account, most of the relevant material within a week, where the case handlers would otherwise have needed months.
Resale price maintenance. Retail prices in online commerce can be captured automatically over long periods. Where a retailer never departs from the manufacturer's recommended price for months, where several retailers always charge the same price, or where several retailers change their prices at the same time, this can be an indication of an infringement. On this basis the Latvian authority opened its first case resting solely on price data it had collected itself. In April 2026 it led to an infringement decision on the fixing of resale prices. The case concerned the distribution of JURA coffee machines: the exclusive Baltic distributor had for years enforced what it called non-binding price recommendations as minimum prices, monitored its retailers' prices and pressed them to comply. Fines of around EUR 512,000 in total were imposed on the distributor and two retailers. The same tool also triggered proceedings against Latvia's largest food retailers. The Czech and Slovak authorities operate comparable price monitoring.
Public procurement
How the screening works. Competition authorities in a number of countries have begun to store procurement data in structured form – bidders, bids, prices, products and the tender documents submitted – and to search it for anomalies using AI. Such anomalies may be the gap between the best and the second-best bid, the dispersion of the losing bids, the rotation of awards among bidders across regions or years, or the question of which bidders regularly bid together or never against each other. Procedures and sectors are pre-sorted by anomaly and then reviewed manually by case handlers.
What it produces. Screening of this kind in procurement procedures is now used in Türkiye, Croatia, Denmark, Switzerland and Spain, among others. In Switzerland, for example, a rotating bidding cartel in road construction was detected without any prior tip-off, solely by analysing tender data, and later also proven in court.
The document trail. Beyond the substance of the bids submitted, authorities sometimes look at how the bid documents came about. What is compared are the metadata of the bids, identical wording and shared errors, and the IP addresses from which bids were submitted. The Brazilian authority has already established shared document origins and shared digital infrastructure among ostensibly independent bidders. On that basis it obtained search warrants against twelve undertakings and, in December 2025, opened formal proceedings against 16 undertakings and 15 individuals; the contracts concerned were estimated at USD 2.2 billion.
Merger control
Unnotified concentrations. Several authorities now actively look for transactions that have not been notified to them. The approaches are similar: state registers are searched for changes in ownership, and press reports and company announcements are monitored and roughly checked against turnover data. The Latvian authority identified 30 cases in this way in 2025 that had not been notified to it and examined whether the standstill obligation had been infringed.
What Austria and the European Commission are doing
The BWB. The Austrian Federal Competition Authority reports that the department for forensics, data analytics and AI set up in 2024 has developed its own platform for reviewing large volumes of digital evidence. The first focus is on analysing e-mail communication and “complex communication networks” (that is, material secured in the course of dawn raids); at a later stage further channels including instant messaging data are to be covered, in so far as that data has been lawfully obtained. Second, a preliminary project running for at least a year is evaluating a future eDiscovery platform, staffed with lawyers, economists, case handlers and forensic experts, and intended to ensure that any future solution meets the requirements of all parts of the authority. Third, a prototype of an internal data analytics platform has been successfully tested; it makes earlier merger decisions searchable by market segment and displays ownership analyses graphically. The BWB has also reported in the past that it systematically evaluates public data sources in order to detect infringements proactively. In addition, the BWB is in discussions with other public bodies on building a database for the analysis of bid data in procurement procedures, in order to detect bid rigging more effectively.
The European Commission. The Commission is one of the global front-runners in the use of screening tools and relies on the analysis of publicly available information to detect cartels. It has already screened thousands of earnings calls algorithmically against some 500 keyword pairs pointing to statements about future pricing and about competitors – such as “price discipline” and “market leader” – and subsequently carried out dawn raids at tyre manufacturers. Around half of its cartel investigations are opened ex officio – a comparatively high figure, indicating active enforcement on the authority's own initiative.
Between the lines: what to take away
The new investigative methods affect businesses in several ways.
The risk of anticompetitive conduct being detected rises. Screening publicly available information, or information already held by the authorities (from procurement databases or register data), with algorithmic or AI-based tools considerably increases the likelihood that anticompetitive conduct is detected. The dawn raids at tyre manufacturers illustrate the point: the Commission carried them out after reviewing thousands of earnings calls had revealed anomalies in that sector.
Assess your risks. Businesses should ask themselves what information about their own market conduct is publicly available, what information they share in public communications, what conclusions can be drawn from it, and whether that can give rise to a competition law risk. The competition authorities – and their AI tools – are reading along. Public communications should be reviewed through a competition law lens before publication, and the employees involved should receive competition law training.
Compliance becomes more important. With a higher risk of detection, the demands on competition law compliance grow. Businesses should ask themselves whether their staff have sufficient compliance knowledge and refresh it regularly. Competition law training should be tailored to sales, procurement, HR and public relations functions, each of which carries its own competition law risk exposure.
The likelihood of incidental findings rises. In dawn raids, competition authorities seize large volumes of data. Searching that data with AI-based tools, alongside manual review by case handlers, increases the likelihood that incidental findings are made beyond the allegation set out in the inspection decision. It is therefore all the more important, during a dawn raid, to ensure that only data covered by the inspection decision is seized – and that the company's internal dawn raid guidelines are up to date.
The use of AI by competition authorities affects every sector – not only undertakings active in sensitive areas, such as those taking part in public tenders. In further blog posts we will look more closely at individual fields in which competition authorities use AI, and show what that means for businesses.
Disclosure: This article was written the way we work. The topic, the analysis, the arguments and the final wording are ours. We use AI as a research and drafting tool along the way. The result is meant to speak for itself.