Europe is the “world’s largest live experiment” in using AI to fight corruption – study

A new paper by Harvard University scientists suggests Europe is running the world’s largest experiment on whether AI can do what institutions couldn’t – fight corruption in the public sector.

The paper, authored by researchers Mark Esposito and Bruno S. Sergi, states that while Europe has lost the AI race to the US and China, it is turning the technology back on its own institutions.

For example, procurement officers, prosecutors, and anti-corruption agencies in Europe deploy graph models, natural language processing pipelines, and risk-scoring systems.

All this amounts to “the world’s largest live experiment” on whether AI can do what institutions couldn’t, on the premise that algorithms cannot be “bribed, intimidated, or invited to dinner.” The study suggests that the early returns are uneven.

For example, Italy’s National Anti-Corruption Authority (ANAC) has run supervised machine learning models on the country’s public procurement database for the past three years.

AI is used to score tenders based on patterns associated with rigged contracting, such as single-bidder awards or unusual cost overruns, and reroute high-risk contracts to human investigators.

A 2024 review credited the authority’s AI system with savings of 10-20% in healthcare procurement alone, translating into €935 million ($1.09 billion) per year.

“AI cannot force authorities to pick up the phone”

The European Public Prosecutor’s Office (EPPO) has built one of the most ambitious AI investigative stacks of any prosecutor’s office in the world, according to the study.

Its toolkit combines network graph analysis to map shell-company structures across jurisdictions, natural-language processing of invoices, and unsupervised anomaly detection of transaction flows, among other functionalities.

The system paid off: the authority ended 2024 with 2,666 active cases covering €24.8 billion in damage to the EU budget, a 38% rise in active casework over 2023.

However, the study emphasizes that more than 70% of EPPO’s 2024 case reports came from private parties, and only 1% from its anti-fraud body, the European Anti-Fraud Office (OLAF).

The authors say this illustrates the institutional ceiling on every AI deployment. While AI can read invoices much faster than human prosecutors, it cannot force OLAF to pick up the phone. Nor can it make a national tax authority share its data with a national police force.

The AI minister may be unconstitutional

In 2025, Albania took an unprecedented step by appointing “Diella” as the country’s “AI minister.”

Prime Minister Edi Rama justified the appointment by saying that “it does not accept coffee and is not intimidated by retaliation” and claimed that the tool has already saved over €300 million ($349 million) in time and costs.

According to the study, the system’s initial procurement workflow is limited to drafting terms of reference, setting eligibility criteria, fixing upper-bound tender prices, and verifying submitted documents. A human procurement officer signs off at each stage.

Have thoughts about this topic? Others do, too. Join them in the discussion.

However, Albania’s opposition challenged Diella’s appointment in court, arguing it was unconstitutional.

Meanwhile, the EU AI Act classifies public-sector decision-making tools as high-risk, requiring human oversight, assessments of the impact on fundamental rights, and EU-level registration.

The study concludes, “​Diella is the sharpest illustration of a problem every government in this story will eventually face. The legal frameworks governing algorithmic authority are still being written in real time, and Albania has decided to write them in costume.”

Unlock exclusive Cybernews content on YouTube.