His responses to these questions were fed into an algorithm that determined how much he would have to pay to access public healthcare. The algorithm, deployed by Kenya’s Social Health Authority, has calculated the public healthcare premiums of millions of Kenyans. In official statements, Kenyan officials have touted the tool, referred to as the “Means Testing Instrument,” as an objective way to ensure that households contribute their fair share in a country where most people work outside of the formal economy and few have official income or tax records. The system does not measure income directly. Instead, it uses machine learning to predict a household’s income based on 43 variables, ranging from the gender of the household head to whether their roof is made of tiles or iron sheets.

Kenya has been part of a long-running global trend in which algorithms are increasingly promoted across the Global Majority as neutral, efficient answers to intractable social problems. In Jordan, an algorithm determines who should receive cash transfers; in Peru, which elderly should receive a pension. The increasing spread and reach of these systems — and the life-changing decisions they make about some of the world’s most vulnerable populations — make journalistic scrutiny essential.

Over the course of more than ten months, Lighthouse Reports, in collaboration with Africa Uncensored and The Guardian, conducted an investigation into how a machine learning model is being used to set the health insurance premiums of millions of Kenyans. By obtaining the training data and variables, we were able to reconstruct the model in order to test its effectiveness and how it would calculate premiums for different kinds of people.

This methodology explains how we approached our analysis. The underlying code and full results have been published to Github.

The Social Health Authority did not respond to a detailed set of questions regarding the findings presented in this methodology.

The post How we investigated Kenya’s AI Means Testing System appeared first on Lighthouse Reports.

To determine who was arrested during the operation, we analyzed more than a million ICE records released by the Deportation Data Project. The data trove contains anonymized individual level ICE records covering arrests, detainers, and detentions from September 2023 to October 2025, in addition to encounters and removals data from September 2023 to July 2025.

By linking these data with bystander video, police dashcam and bodycam footage, and highway patrol incident reports, we reconstructed the anatomy of the operation carried out by ICE and Tennessee Highway Patrol. This methodology shows how we used these data sources to identify people arrested during the operation, to test ICE’s claims against what really happened and to tie individual stories to regional and national trends in immigration enforcement under the Trump administration.

Highlights of our findings include:

● Arrests in Tennessee have surged under Trump, even as ICE has arrested a smaller percentage of people convicted of a crime, mirroring national patterns.
Community ICE arrests, which have been criticized nationally, have a far lower than average success rate in apprehending people with criminal convictions.

● Arrests in Tennessee spiked in early May 2025 when ICE collaborated with Tennessee Highway Patrol to carry out traffic stops and arrest undocumented drivers.

● DHS lied about who was arrested during the traffic operation, falsely giving the impression that it led to the arrest of a rapist and a drug dealer. In addition, it is very likely that DHS overstated the number of arrests it made.

● We identified who was arrested in the operation. We can show that few of them had serious criminal backgrounds, mirroring regional and national patterns.

● We traced ICE arrestees through a succession of detention facilities. Most were quickly transferred to Louisiana and Texas, places with very low access to legal support, before being deported.

All of the code used for analysis is available on our Github.

The post How We Unpacked Tennessee’s Cooperation with ICE appeared first on Lighthouse Reports.

The data came from a little-known surveillance company called First Wap. Headquartered in Jakarta but run by a group of European executives, First Wap has quietly built a phone tracking empire spanning the globe.

There have been leaks of telecom network targeting data in the past (some of which Lighthouse has written about). But none of them has included this amount of successful targeting of individual phone numbers.

The archive was a starting point for Surveillance Secrets, a collaboration between Lighthouse Reports, paper trail media and 12 other partners that lifts the lid on First Wap’s activities. The team found material inside the archive for dozens of stories, including how the company’s tracking tech was used against Rwandan dissidents targeted in an assassination campaign, a journalist investigating corruption in the Vatican, and a businessman being investigated for compromising material.

Getting to those stories took months of analyzing 1.5 million rows of obscure telecom data. Unlike top-tier spyware firms, such as the notorious NSO Group, phone-tracking firms like First Wap have flown under the radar. Standard tools used in spyware investigations — such as device forensics — were unavailable to us; there was no blueprint examining a firm such as First Wap. In this technical explainer, we explain how we approached the data to understand the company’s operations.

The post How First Wap Tracks Phones Around the World appeared first on Lighthouse Reports.

Governments and companies deploying these systems often show little regard for the biases they perpetrate against vulnerable groups. The city of Rotterdam told us that it had never run code designed to test whether its model was disproportionately flagging vulnerable groups. France’s social security agency, CNAF, confirmed to us that it had never audited its model for bias.

In January of 2023, three months before publishing our investigation into Rotterdam’s risk scoring algorithm, we sent a public records request to the city of Amsterdam, one of Europe’s most progressive capitals. Among other things, we asked for documents, code and data relating to a similar system Amsterdam had been developing. Given that we had fought over a year to obtain these types of material in other investigations, we were surprised when the city immediately complied with our request.

The materials disclosed by the city were related to a machine learning model it was developing in order to predict which of the city’s residents were most likely to have submitted an incorrect application for welfare. At the time of our public records request, the model was still in development. The overall development goal for the model, according to the city’s internal documentation, was to have fewer welfare applicants investigated, but a higher share of those investigated rejected. Internal documents also emphasized two other aims: avoid bias against vulnerable groups, and outperform human caseworkers.

After reading through the documentation, it quickly became clear to us why the city had been so forthcoming. The city had gone to significant lengths in order to develop a model that was transparent and treated vulnerable groups fairly.

We wanted to investigate whether the city of Amsterdam had succeeded in developing a fair model. Over the course of eight months, we ran a series of our own tests on the model and data containing real world outcomes for people flagged as suspicious by the model. When it came to the outcome data, the city ran our own tests locally and returned aggregate results in order to comply with European data protection laws.

The past five years have seen a number of regulatory efforts to reign in harmful uses of AI. The EU AI Act, passed in 2024, will require AI systems deemed “high risk” to be registered with the European Commission. In New York City, employers have been prohibited from screening employees using AI without conducting bias audits since July 2023. Meanwhile, the private sector, academics and multilateral institutions have produced a number of responsible AI frameworks. In taking on this investigation, we wanted to look ahead and understand the thorny reality of building a fair AI tool that makes consequential decisions about people’s lives.

The code and data underlying our analysis can be found on our Github.

The post How we investigated Amsterdam’s attempt to build a ‘fair’ fraud detection model appeared first on Lighthouse Reports.

Key Findings

– More drowning deaths than previously reported. According to our data, at least 1,107 people drowned crossing the Rio Grande between Texas and Mexico between 2017 and 2023, a figure substantially higher than previously documented.

– Deaths peaked in 2021-2022, years when the number of people crossing was rising and Texas tried to seal its border with Mexico in an initiative called Operation Lone Star.

– The deadliest stretch of the Rio Grande between 2017-2023 was between the Mexican state of Coahuila and the Texas counties of Kinney, Val Verde and Maverick, which includes the city of Eagle Pass, which has been described as “ground zero” for Texas’ Operation Lone Star.

– In 2022 and 2023 the river was more deadly for women who made up one in five drownings. In 2023, one in ten deaths were of children. There were also more drowning victims from countries outside of Mexico and Central America in those years.

– Incomplete official data in both the US and Mexico leaves hundreds of deaths uncounted. No single agency in Mexico is comprehensively documenting migration-related deaths. Meanwhile, our data shows that the U.S. Customs and Border Protection (CBP), the U.S. agency mandated to document migrant deaths, has been severely undercounting drownings in the Rio Grande. The CBP recorded 498 water-related deaths in Texas and New Mexico from 2017 to 2023 while our investigation documented 858 migrant drownings in Texas alone during that time.

The post Rio Grande Methodology appeared first on Lighthouse Reports.

The post How we investigated Sweden’s Suspicion Machine appeared first on Lighthouse Reports.

The post QAnon Methodology appeared first on Lighthouse Reports.

The post Brain Waste Methodology appeared first on Lighthouse Reports.