Europe may have lost the global artificial intelligence race to Silicon Valley and China, but across the continent a quieter revolution is underway — one that could fundamentally reshape the fight against corruption.
From Rome to Kyiv, Brussels to Tirana, governments and prosecutors are deploying artificial intelligence not to build the next ChatGPT, but to expose rigged contracts, identify shell-company networks, and detect suspicious procurement patterns hidden inside mountains of public data.
The promise sounds almost utopian: algorithms cannot be bribed, intimidated, blackmailed, or invited to expensive dinners.
Yet the reality emerging across Europe is far more nuanced. AI is proving remarkably effective at identifying corruption — but only where institutions are willing to act on the evidence.
A new analysis by academics Mark Esposito and Bruno S. Sergi argues that Europe has become “the world’s largest live experiment” in using artificial intelligence to combat public-sector corruption.
Read their full analysis here:
https://www.eureporter.co/wp-content/uploads/2026/05/AI-and-Corruption.pdf
Their conclusion is both encouraging and uncomfortable: AI is not replacing human accountability — it is exposing where accountability already exists, and where it does not.
Italy’s Quiet Success Story
Italy may be Europe’s most unexpected AI anti-corruption success.
For decades, Italian public procurement was synonymous with scandal, patronage, and opaque political deals. Today, however, Italy’s National Anti-Corruption Authority (ANAC) is running sophisticated machine-learning systems across national procurement databases to detect suspicious contracting behaviour.
The AI models analyse more than seventy risk indicators linked to corruption patterns, including:
- single-bid tenders,
- discretionary awards,
- unusual cost overruns,
- last-minute specification changes,
- and irregular procurement procedures.
High-risk contracts are then passed to human investigators.
The results have been dramatic. According to a 2024 review by the Open Contracting Partnership, the system is estimated to save between 10% and 20% annually in healthcare procurement alone — roughly €935 million every year.
Importantly, Italy’s transformation has happened without political theatre or flashy announcements.
The success came from something less glamorous:
clean datasets, competent regulators, and institutions willing to let technology work quietly in the background.
As the report notes, “the most consequential anti-corruption AI in Europe is the kind nobody photographs.”
Brussels: Powerful Technology, Weak Coordination
At EU level, the picture becomes more complicated.
The European Public Prosecutor’s Office (EPPO), operational since 2021, has developed one of the world’s most advanced AI-supported investigative systems.
Its technology stack includes:
- network-graph analysis to uncover shell-company structures,
- multilingual natural-language processing across 24 EU languages,
- anomaly detection systems analysing financial transactions,
- automated analysis of invoices, contracts and communications.
The scale is extraordinary.
In 2024 alone, EPPO opened 1,504 new investigations involving an estimated €13.07 billion in damages to the EU budget. By year-end, the office had 2,666 active investigations worth nearly €25 billion.
Yet behind these impressive figures lies a troubling institutional weakness.
More than 70% of EPPO case referrals came from whistleblowers, journalists, private contractors and citizens — while only around 1% came from OLAF, the EU’s own anti-fraud office that was specifically designed to support such investigations.
The problem, therefore, is not the AI.
It is the institutions.
Artificial intelligence can scan millions of records in seconds, but it cannot force agencies to cooperate, share information, or pursue politically sensitive cases.
As the report bluntly observes, “the technology was never the bottleneck. The plumbing was.”
Ukraine: The Unexpected Pioneer
Perhaps the most surprising European AI success story is Ukraine.
Following the 2014 Revolution of Dignity, Ukrainian reformers launched Prozorro, an open-source e-procurement platform designed to bring transparency to public spending.
Its companion platform, DOZORRO, became one of the world’s first systems to apply machine learning directly to corruption-risk detection in procurement.
The AI system flags suspicious tenders for review by civil-society monitors, procurement experts, and investigators.
The results have been impressive:
- 26% more cases of unjustified contract awards identified,
- 37% more improper disqualifications detected,
- and nearly $6 billion in public funds reportedly saved since 2017.
Remarkably, much of this progress occurred during wartime.
Ukraine’s experience has become an important model for European policymakers because it combines three elements often missing elsewhere:
- AI-driven risk detection,
- human oversight,
- and public transparency.
The system does not allow algorithms to operate in secret. Civil society can independently verify the same data being analysed by the state.
That balance between automation and democratic oversight may prove crucial for Europe’s future regulatory direction.
Albania’s AI Minister
If Italy represents quiet efficiency, Albania represents spectacle.
In September 2025, Albania appointed “Diella,” an AI-powered chatbot, to oversee parts of the country’s public procurement system.
Presented visually as a woman wearing traditional Albanian dress, Diella became what Prime Minister Edi Rama described as the world’s first “AI minister.”
The government claims the system has already generated savings of more than €300 million through efficiency gains and reduced corruption risks.
Behind the headlines, however, the technology itself is relatively modest.
Built with support from Microsoft and based on OpenAI’s GPT architecture, the system currently assists with:
- drafting procurement documents,
- setting eligibility criteria,
- verifying submissions,
- and estimating tender pricing.
Human officials still approve all final decisions.
Nevertheless, Diella has triggered a serious constitutional debate.
Opposition politicians argue that Albanian law requires ministers to be “mentally competent citizens,” raising unprecedented legal questions about algorithmic authority and democratic governance.
The controversy also highlights broader tensions emerging under the EU AI Act, which classifies many public-sector AI systems as “high-risk” technologies requiring strict human oversight and transparency safeguards.
Europe’s Real Challenge
Beyond the headlines, a broader European infrastructure is rapidly emerging.
Projects such as DATACROS III — backed by the EU Internal Security Fund — now use graph-based machine learning to analyse beneficial ownership structures across more than 400 million companies worldwide.
Its systems identify:
- opaque ownership chains,
- politically exposed persons hidden behind nominees,
- circular shareholdings linked to money laundering,
- and sanctions evasion networks.
Meanwhile, EU-backed training programmes are preparing prosecutors, auditors and procurement officers across the Balkans and Eastern Europe to work with AI-assisted investigative tools.
But the report warns that major risks remain.
Among the biggest concerns are:
- lack of explainability in AI decisions,
- protection of highly sensitive financial data,
- and the growing question of whether algorithmic governance is fully compatible with European constitutional principles.
The Next Five Years
Europe’s anti-corruption AI experiment is no longer theoretical.
The systems are already operating.
The investigations are already underway.
The political consequences are beginning to emerge.
The real question is no longer whether AI can detect corruption.
Increasingly, it can.
The question is whether governments are willing to confront what the algorithms reveal — especially when the targets are politically connected individuals, institutions, or ruling parties themselves.
That is the true test now facing Brussels, Rome, Kyiv and Tirana alike.
And it is a test that artificial intelligence cannot answer on its own.
