AI systems are now being used in European procurement processes to tackle corruption. Mark Esposito and Bruno S. Sergi write that the political question is no longer whether this technology works, but who decides when it is allowed to work.
Europe lost the AI race to Silicon Valley and Shenzhen years ago. What it is doing now is politically more interesting: using AI-powered tools to detect, prevent and investigate corruption.
Procurement officers and prosecutors are deploying graph models and risk-scoring to counter the closed-bidding practices that elected officials and their networks have long relied on. The targets are the patronage routines of European public administration.
Algorithms cannot be bribed or invited to dinner. That framing enables governments to adopt the systems while avoiding the admission that their own enforcement apparatus was the bottleneck the technology was built to route around. But AI can only catch corruption if a human acts on what it flags. The political question now is not whether AI anti-corruption systems work, but who decides when they are allowed to work.
AI anti-corruption systems and enlargement
The backdrop to all of this is the EU’s revived enlargement process. Enlargement Commissioner Marta Kos has called a new wave of accessions by 2030 a realistic prospect.
Montenegro is the frontrunner, with the country targeting full membership by 2028. Albania has opened all 33 negotiating chapters in just over a year and aims to close them by 2027 for accession by 2030. Ukraine and Moldova completed bilateral screening in September 2025 and have signalled their intent to close negotiations by the end of 2028, though Hungary has blocked the formal opening of clusters.
Procurement integrity is the chapter where every one of those bids is most exposed, because EU pre-accession funds – the Growth Plan for the Western Balkans, the Moldova Growth Plan and the €50 billion Ukraine Facility – will start flowing in volume before institutional habits change. That is why anti-corruption AI has moved from a domestic reform topic into a foreign-policy instrument.
Brussels needs candidates who can demonstrate enforcement capacity that can be verified from the outside, and candidates need a story they can take to voters about who is actually being constrained. The systems are arriving faster than the institutions designed to govern them, and the political asymmetry is unmistakable: every government wants the legitimacy that comes with deploying anti-corruption AI, but nobody wants to be the first government those systems flag.
Brussels – the prosecutor with no phone
At the EU level, the design problem becomes visible. The European Public Prosecutor’s Office (EPPO), operational since 2021, is the first EU body empowered to prosecute crimes against the Union’s financial interests directly, bypassing national prosecutors. Its toolkit combines graph analysis of shell-company structures, natural-language processing of invoices in 24 languages and anomaly detection on transaction flows. In 2024 it opened 1,504 investigations representing 13.07 billion euros in estimated damage and obtained 2.42 billion euros in freezing orders.
Then comes the bureaucratic-politics problem. More than 70 per cent of EPPO’s 2024 case reports came from private parties, including journalists, whistleblowers and losing bidders. Only 1 per cent came from OLAF, the EU’s anti-fraud body, which predates EPPO, reports to the Commission and has every incentive to defend its referral monopoly. AI cannot resolve a turf war between two Brussels agencies. The technology was never a bottleneck. The institutional design was.
Kyiv – the model nobody expected to outlast a war
Ukraine inverts the assumption that anti-corruption infrastructure follows state capacity. After the 2014 Revolution of Dignity, with the central state visibly weakened, activists and reformist officials built Prozorro, an open-source e-procurement system that handled 3.6 million tenders in 2024. Its civic-monitoring counterpart, DOZORRO, was the first public-procurement platform to use machine learning for corruption-risk detection, half a decade before the EU AI Act, because reformers did not trust the post-Maidan state to police itself.
The political innovation is the routing. High-risk contracts are first brought to the attention of civil-society monitors, who decide whether to escalate them to auditors or law enforcement. Harvard Kennedy School credits Prozorro with helping save Ukraine nearly 6 billion dollars since 2017, most of it overlapping a full-scale war. The design that Brussels now codifies as best practice was developed in Kyiv eight years earlier and embedded civil society into the enforcement chain. A constitutional move, not a technical one.
Albania’s AI-generated minister
In September 2025 Albania gave a cabinet seat to a chatbot. Diella, an AI “minister” rendered onscreen in traditional Zadrimë dress, formally oversees public procurement. Stripped of the costume, she drafts terms of reference, sets eligibility criteria and verifies documents, with a human signing off: roughly what Italy’s independent anti-corruption regulator ANAC has done for three years. The difference is the chain of command. ANAC reports to no minister. Diella reports to the prime minister who appointed her.
The constitutional fight is the point. Albania’s opposition has challenged the appointment, arguing that ministers must be “mentally competent citizens”. It sounds like a stunt, but it is the cleanest articulation yet of the question every European democracy will face. The deeper issue is that Diella lets a head of government claim algorithmic neutrality for decisions he still controls. The politicians most eager to invoke algorithmic authority are not always the ones most willing to be bound by it.
What the next five years will test
Europe’s anti-corruption AI moment will be judged on two questions. First, whether the prosecutors and monitors using these tools are protected when flagged contracts point at the people who control their budgets. Second, whether the politicians who commissioned the systems are prepared to be the first targets when the models work as advertised. The technology is no longer the variable. The political will to live with its findings is.
Note: This article gives the views of the authors, not the position of LSE European Politics or the London School of Economics.
Image credit: Khakimullin Aleksandr provided by Shutterstock.
Discussion about this post