Ghost Busters

Frontier Technologies in the Fight Against Corruption

Billions spent. Nothing built.

In the Philippines, over 400 "ghost projects" have been uncovered in flood-control infrastructure alone. Billions of pesos were disbursed. Many communities received nothing. These aren't bureaucratic failures. They're thefts of public money and the trust that functional government depends on.

Across Asia, Latin America, and beyond, governments and civil society are deploying AI, satellite imagery, open data platforms, and citizen reporting tools to detect irregularities, track project delivery, and build evidence for prosecution. The results are uneven. Context matters enormously. But the trajectories are real.

Through research and horizon scanning, FT Hub and partners have explored the tech vs corruption landscape. Our report profiles 6 international case studies, 8 technologies, and 5 trends.

The anti-corruption chain

Corruption in public infrastructure doesn't happen in one place. It happens at multiple points along a chain, from the moment a project is specified through to the moment funds are released for work that may never have been done. The technologies described in this paper each address different points in that chain.

    
    1
    Prevent manipulation
    before award

    2
    Detect risk during
    procurement

    3
    Verify delivery
    after award

    4
    Enable oversight
    and reporting

    5
    Strengthen
    enforcement

  1
Prevent manipulation before award
2
Detect risk during procurement
3
Verify delivery after award
4
Enable oversight and reporting
5
Strengthen enforcement

Before asking where the field is heading, it's worth understanding the technology families at work. What each one can reliably do today, where its limits are, and where it's heading over the next five years.

inds suspicious patterns in large procurement datasets that human auditors can't feasibly review. Flags unusual specifications, improbable win rates, and systematic contract amendment patterns. Used for triage, not automated judgement.
In five years
LLMs will read the actual text of tender specifications and assess whether requirements appear designed to exclude competition. Unsupervised anomaly detection will work even where historical corruption cases are rare or suppressed.
The only technology that can independently verify physical reality without human presence. Confirms whether a project was actually built, in the right location, at approximately the right scale. Can't easily verify whether it meets contractual specifications.
In five years
Falling imagery costs, improved computer vision, and multi-sensor integration (LiDAR, multispectral) could enable remote assessment of approximate dimensions, material consistency, and structural integrity from orbit.
The enabler layer that makes everything else possible. Standardised, machine-readable procurement data is the prerequisite for every analytical technology. The persistent gap is between mandate and reality: passing a law and having clean, timely data published are different things.
In five years
The shift from open procurement data to linked open government data, connecting procurement records to company registries, beneficial ownership databases, asset declarations, and budget execution systems, creates detection possibilities impossible with procurement data alone.
Maps connections between companies, individuals, government bodies, and contracts to reveal conflict-of-interest structures invisible in transaction-level data. Requires clean entity identifiers. Network proximity is not proof of corruption, and systems need strong due process safeguards.
In five years
Integration of beneficial ownership data and LLM extraction of relationship data from unstructured sources (news, court records) could flag, in near real time, any tender where a bidder's beneficial owner has a documentable relationship with an evaluating official.
Distributes detection capability across a population that no government institution can match for scale or local knowledge. Citizens can interrogate procurement data, flag incomplete projects, and submit evidence. The persistent limit is sustainability: civic tech is frequently under-resourced and politically vulnerable.
In five years
AI-assisted tools in local languages, integrated with satellite verification, could create a powerful pipeline: a citizen flags a stalled project, imagery confirms no construction activity, and the combined evidence package is automatically routed to an oversight body.
Creates tamper-evident, append-only records. Its honest current use is narrow: preventing retroactive record manipulation. Most procurement corruption doesn't depend on altering records after the fact, it depends on making the right decisions before the record is created.
In five years
Most defensible role is likely as an integrity layer for specific high-value, high-risk transactions such as donor-funded projects, and for tracking physical supply chains to confirm that materials purchased were actually delivered to site.

What the tech can actually do

What has been tried internationally?

The following cases span different technologies, different institutional contexts, and different parts of the world. None is a blueprint. Together, they describe what works, what doesn't, and what conditions tend to make the difference.

Thailand

When a viral post changed national procurement

A villager's discovery of ornate lamp posts on a vacant plot, shared anonymously via a LINE chatbot and anti-corruption Facebook page, triggered a nationwide investigation into contracts dating back to 2013. ACT AI aggregates 31 million procurement records with rule-based red flag detection. The NACC ultimately charged more than 40 people.

Open data alone is not enough. Civic engagement, press freedom, and effective investigative authorities are all crucial.

Ukraine

Building "everyone can see everything" into procurement

Post-revolution, Ukraine rebuilt procurement transparency with Prozorro (all data open by default) and Dozorro (independent civic oversight operated by Transparency International Ukraine). Corruption reduced by 25%; savings of $2.7M daily. The design principle: state data infrastructure, independently operated monitoring layer.

Digitising procurement cuts discretion, but its gains depend on pairing the open backbone with an active civic escalation layer.

Brazil

An AI robot that reads 190,000 contracts a year

Brazil's ALICE Robot continuously screens federal procurement documents for overpricing, short timelines, and specifications that may restrict competition. In 2023, its alerts triggered 203 audit jobs covering R$27 billion in procurement. ALICE flags risk for human auditors to act on — it does not automate judgement.

AI can materially strengthen procurement integrity when used as a triage for human auditors, not a replacement for them.

South Korea

End-to-end digitisation that closed the discretionary backdoor

When bid clustering revealed systematic information leakage in Korean public construction contracts, South Korea replaced fragmented paper processes with KONEPS: a single digital procurement backbone covering the full lifecycle. But digitisation also created new fraud surfaces — scammers exploited public contracting data to impersonate government bodies.

End-to-end digitisation can shrink corruption space, but also shifts the threat landscape. Strong identity controls are essential.

Chile

An observatory that catches irregularities while they are still live

ChileCompra's Public Contracting Observatory runs automated monitoring across its national platform, prioritising higher-risk transactions for follow-up while a procurement is still live. Now integrating LLMs for predictive anomaly detection across tender processes and confidential complaints.

AI increases the reach of oversight when used for early warning. Its gains depend on clean data, clear escalation pathways, and human capacity to follow through.

Colombia

Testing whether blockchain can harden the tender record

Following a major corruption scandal, Colombia tested blockchain architecture in public school meals procurement, creating tamper-evident bidding records with real-time auditability. Developed with Procuraduría General, C4IR Colombia, and Colombia Compra Eficiente. WEF described the outcome as a basis for cautious optimism.

Blockchain solves a narrow problem. Most corruption happens before the record is created, not in altering it afterwards.

Five directions of travel

So where is this heading? These aren't predictions. They're trajectories already visible in pilots, deployments, and research, that will shape what anti-corruption technology looks like in the Philippines and elsewhere within the next five years.

The traditional anti-corruption model is retrospective: something goes wrong, evidence accumulates, an investigation begins. The emerging model intervenes before a corrupt decision is finalised. Systems are being designed to flag irregularities while a tender is still live, alert officials before a contract is signed, and create friction at the moment corruption is most stoppable.
Paraguay's procurement authority issues real-time alerts inside its e-procurement platform, mapped to specific legal violations, at the exact decision points where officials can still act. Brazil's ALICE system has moved from periodic audit to daily automated screening of hundreds of thousands of processes. The Dominican Republic has linked red-flag monitoring directly to supplier debarment, closing the loop between detection and consequence.
For decades, anti-corruption efforts focused on the moment of award. But ghost projects don't fail at the tender stage, they fail at delivery. The frontier has shifted to tracking a project from pipeline stage through physical construction, and eventually tying verified delivery directly to payment release. When every stage of a project leaves a digital trail, ghost infrastructure becomes structurally harder to sustain.
Ukraine's DREAM platform tracks over 12,000 reconstruction projects end-to-end across 87% of local governments. The World Bank's GEMS initiative has deployed GPS, photo verification, and satellite imagery across 1,000+ projects in 100+ countries. Satellite-based ML models are being tested to assess road quality over time from orbit, enabling auditors to cross-check contractor progress claims without setting foot on site.
The most resilient anti-corruption systems in this review share a design principle: they don't rely on government alone to police government. The pattern that works is hybrid: state-managed data infrastructure that is open by default, combined with independently-operated monitoring by civil society, journalists, and citizens. Neither alone is sufficient. Governments can suppress or defund official tools when politically inconvenient. Civil society can't act without access to the underlying data.
In the Philippines, BetterGov.ph volunteers have already connected PhilGEPS, DIME, budget, flood control, and political dynasty data into a publicly accessible analytical platform. Bisto.ph has secured a formal DPWH partnership for citizen ground-level inspection of 250,000 infrastructure projects nationwide. The architecture that works is the one where data is legally mandated to be public, and the monitoring layer is structurally independent.
Until recently, building a procurement risk analytics system required significant custom development. That constraint is dissolving. New open-source infrastructure means the capability to run red-flag analytics, network analysis, and risk scoring is increasingly available off the shelf. The binding constraint is no longer "can we build this?" It's "do we have clean enough data, sufficient institutional will, and a credible enforcement hook at the end?"
The Open Contracting Partnership's Cardinal library automates red-flag calculation for any government publishing data in the open contracting standard, deployable by a civil society organisation without major technical investment. In the Philippines, the 2024 New Government Procurement Act has created the open contracting data mandate that makes this possible.
Most tools in this landscape stop at the same point: a flag, a dashboard, an alert. The emerging frontier is systems designed not just to detect risk but to package evidence for enforcement, building the chain from algorithmic signal to prosecutorial action. This is the least developed part of the landscape, and the most consequential.
Kazakhstan's redflags.ai routes procurement anomaly alerts directly to prosecutors. The World Bank's GRAS deployment in Brazil surfaced over 850 suppliers showing signs of collusion, 450 likely strawmen, and 500 cases of companies owned by public servants receiving contracts from their own agencies. BetterGov.ph has built a political dynasty mapping tool cross-referencing procurement awards against family networks across 82 Philippine provinces.

Read the report in full

This page covers the highlights. The full paper includes detailed case write-ups, a complete technology landscape assessment, methodology notes, scoring across eight dimensions of Philippine fit, and exposes the 5 trends to adversarial critique from 3 AI-generated personas. Fill in the form to download it.