Spain scans 6 million contracts for cartels

Imagine a machine that remembers every government tender ever published, every firm that bid, every price offered, every subcontractor hired afterwards. Now imagine it scanning for the same patterns a human investigator would spot if they had a thousand years to read the files. That is what Spain's competition authority is building.

The Comisión Nacional de los Mercados y la Competencia (CNMC) is developing an AI system called Atenea to scan public procurement data for signs of bid-rigging and collusion (La Vanguardia). The tool draws on a database of roughly six million contracts and is designed to flag suspicious tenders for human investigators, not to prove guilt on its own.

What the Machine Looks For

Cartels in public procurement leave fingerprints. Companies taking turns to win. Losers submitting bids that are suspiciously high. A firm that always loses but then gets the subcontract. The OECD's bid-rigging guidance has catalogued these signals for years. The problem was never knowing what to look for. It was looking everywhere at once.

A local official sees one tender. Atenea can see repeated behaviour across buyers, regions, sectors and years. It reportedly works in layers. First, it links companies that appear under different names, the way a single firm might bid through three subsidiaries. Then it groups suspicious patterns: do the same two companies keep showing up together? Finally, it asks whether a given market behaves differently from what genuine competition would predict.

The system is reported to evolve from an earlier, narrower CNMC screening tool, expanding into open markets and a newer concern: algorithmic collusion. That is when companies' pricing software learns to match competitors' moves without any human ever picking up the phone, the way shops on the same street might silently stop undercutting one another.

Why Spain Has Reason to Hurry

This is not a solution looking for a problem. Last month, Spain's Supreme Court confirmed a €13.5 million fine against Indra for participation in a cartel rigging public IT-services tenders, with conduct spanning 2005 to 2015 (Cinco Días). The practices included cover bids, competitors refraining from bidding, and advance knowledge of tender details through insider contacts. Indra was attributed more than €324 million in contracts awarded through the affected arrangements (elDiario.es).

Those are precisely the patterns a machine could detect far earlier than a decade-long investigation: the same firms appearing together, winners and losers swapping roles, a losing bidder quietly reappearing as subcontractor.

The Accountability Test

Spain is part of a wider European shift toward machine-assisted enforcement. Germany's proposed 12th amendment to its competition law would enable systematic procurement screening (BBH Blog). Italy's anti-corruption authority ANAC already publishes tender data that creates raw material for similar analysis (ANAC).

The harder question is what happens when the algorithm decides who gets investigated. The Netherlands learned this painfully when a court struck down SyRI, a government risk-profiling tool, for violating privacy rights (Rechtspraak). The Dutch response was to build a national register where 1,495 algorithm descriptions from 515 organisations are now publicly visible (Algoritmeregister). The principle behind it: if a state algorithm shapes who gets scrutinised, citizens need to know it exists.

Academic evidence supports screening's promise but warns against overselling. A CEPR study of Swedish pharmaceutical auctions found patterns consistent with collusion alongside higher prices, but results like those are market-specific. They do not prove universal AI detectors work everywhere. False positives remain the core risk: specialist contracts naturally attract few bidders, common input costs can make prices move together, and joint bidding can let small firms access contracts they could not handle alone.

Atenea is reported as still in internal development, with deployment expected in coming months. No public technical specification has surfaced. The CNMC has not published the system's architecture, error rates or governance rules.

The tool's legitimacy will depend less on how clever the algorithm is and more on whether its alerts can be challenged, its logic audited, and its recommendations overruled. Spain is testing whether enforcement can move earlier in the cycle, spotting collusion before a whistleblower or a decade-long court case surfaces it. If it works with proper safeguards, taxpayers and honest firms both benefit. If it operates as a black box, it risks becoming the kind of unchallengeable algorithmic authority that European courts have already rejected elsewhere.