Electrocutions, Intimidation and Wage Theft Haunt Europe’s Data Center Workers

A pedestrian passes a downtown mural honoring World War II veterans in California’s 22nd Congressional District on October 22, 2026 in Bakersfield, California, a competitive southern Central Valley district with a majority Latino population. (Photo by Mario Tama/Getty Images)

What Latino Voters Can Teach Washington About AI Governance

A digital collage combining a historical painting of a miner with a large grid of small online images. At the center, a muscular miner works underground with a heavy drill inside a dark tunnel. To the left and along the bottom, hundreds of tiny square images form a dense mosaic, suggesting scraped internet content. A bright beam of light connects the image grid to the miner’s drill. On the right, a block of colorful digital noise contrasts with the detailed painting, reinforcing the link between physical mining and data extraction.

Primitive Accumulation by Daniela Zampieri / 

What Training Dataset Transparency Mandates Hide

Demonstrators wear masks depicting Meta CEO Mark Zuckerberg, Brazilian presidential candidate Flavio Bolsonaro and US President Donald Trump during a protest outside Meta's headquarters in Sao Paulo, Brazil, on October 19, 2026. The protest was called over allegations that restrictions on Instagram affected the campaign of Brazilian President Luiz Inacio Lula da Silva ahead of the presidential runoff election. Meta attributed the reported issue to a technical failure and denied political bias. (Photo by Allison Sales/NurPhoto via AP)

Brazil Shows Platform Illiberalism is Big Tech’s New Normal

Grok app icon is seen next to the Meta AI and Claude app icons on a smartphone home screen in this illustration photo taken on October 27, 2026. (Photo by Matteo Della Torre/NurPhoto via AP)

The Solution to AI’s Alignment Problem is Precise Guardrails on How We Use It

A stock market chart with lines in different colors on a dark screen with lines trending downwards to the right. 

Chart on a dark screen by Maxim Hopman on 

Will Prediction Markets Affect Democracy? Take a Bet

Canada's Prime Minister Mark Carney, front left, speaks with European Commission President Ursula von der Leyen at the European Parliament in Strasbourg, eastern France, Sept 17, 2026. (AP Photo/Pascal Bastien, File)

Canada and Europe Need a Way to See the AI They Cannot Access

A surreal collage depicting a dishevelled pile of items arranged like a traditional vanitas painting. To the right is a vintage computer monitor displaying the OpenAI logo. Objects representing various instances of AI “slop” litter the scene, such as an idol of Jesus with a shrimp replacing its head, and an anthropomorphised cappuccino cup with cartoonish eyes and lips.

Slop Vanitas II by Georgina Leeuwendal / 

Creating a New Internet Bargain to Preserve Fair Use and Protect the Commons

A close up of a spider in its web. Valeriia Neganova/Unsplash

A close up of a spider in its web. Valeriia Neganova/

Breaking the Open Web Won’t Stop the Bad Bots

Prime Minister Andy Burnham (left) meets with US President Donald Trump during his visit to the US to attend the United Nations General Assembly on Sept. 22, 2026, in New York. Photo: Toby Melville/PA Wire POLITICS (Press Association via AP Images)

The UK-US Tech Deal Exposes the Limits of Britain's AI Ambitions

A crowd of supporters surround President Lula holding a microphone at a podium draped with the Brazilian flag.

President Luiz Inacio Lula da Silva delivers a victory speech after winning the presidential runoff in Brazil, Oct. 25, 2026. (AP Photo/Andre Penner)

Lula’s Win Signals Fight to Regulate Social Media Will Escalate in Brazil

A data center is under construction in Loudoun County, Virginia on October 7, 2026. The region houses the world's largest concentration of data centers, making Virginia the nation's digital capital. (Photo by Olivier Douliery/Sipa via AP Images)

What to Do About the Coming Glut of AI Data Centers

Tesla and SpaceX CEO Elon Musk receives the National Medal of Science from US President Donald Trump during the Golden Age of American Innovation Summit at the Institute of Peace on October 08, 2026 in Washington, DC. (Photo by Win McNamee/Getty Images)

‘The Greatest Capitalist In History’? A Review of Alex Gibney’s ‘Musk’

Secretary of Defense Pete Hegseth gives a State of the Force address at Marine Corps Base Quantico, Wednesday, Sept. 30, 2026 in Quantico, Va. (Alex Wong/Pool Photo via AP)

The Pentagon Has a Big Tech Problem

An entrance driveway to a hospital with a sign pointing to Emergency, Laboratory, and Radiology. A person is walking past from the left, and another person is leaving the hospital on a motorbike. 

A street view of The Nairobi Hospital in Kenya on Oct. 6, 2026 (AP Photo/Andrew Kasuku)

Without an Exit Plan, AI Partnerships in Africa Risk Becoming the New Tied Aid

Voters line up at a polling station during the presidential runoff in Brasilia, Brazil, Sunday, Oct. 25, 2026. (AP Photo/Eraldo Peres)

How to Reclaim the Future from Strongmen and Oligarchs

The Edge Dialogues with Dr. Alondra Nelson

Demonstrators holding a banner that reads in Portuguese "Brazil decides about Brazil" rally against Meta over allegations of interference in Brazil's presidential election, outside Meta's headquarters in Sao Paulo, Monday, Oct. 19, 2026. (AP Photo/Andre Penner)

On Eve of Vote, Brazil’s Federal Prosecution Office Demands Answers from Meta

Demonstrators wear masks depicting Meta CEO Mark Zuckerberg, Brazilian presidential candidate Flavio Bolsonaro and US President Donald Trump during a protest outside Meta's headquarters in Sao Paulo, Brazil, on October 19, 2026. (Photo by Allison Sales/NurPhoto via AP)

In Brazil’s Runoff, Lula Faces Bolsonaro, Trump—and Big Tech

A grey toned abstract photo of wavy mirror reflections of a building seen from many different angles. 

LLMs Are Still Making Election Deepfakes Just Days Before the US Midterms

A five-month investigation by Investigate Europe in partnership with Tech Policy Press and media partners in nine countries reveals the harsh realities facing some working on the continent's vast AI infrastructure developments.

Sandu Dumitrescu* checked his watch. He was winding down the last 15 minutes of his shift at the Google data center site in Saint-Ghislain, Belgium, in March, when the call came in. He had just finished performing electrical checks on the site’s hall fans when his manager communicated a final request: could he run two quick tests right before he left? The 34-year-old electrical tester says he had to make a decision: he had already worked almost 10 hours. Yet in the highly pressurized environment of GBL-6, Google’s sixth data center in the vast €5 billion campus, such last-minute requests were normal.

“I had no choice,” Sandu says. “People have been fired after saying no.” He believed his Estonian-based supervisors on the Google site, where Irish company Dornan Engineering is the main electrical subcontractor, would not take a refusal well. Relying on his week-by-week contract and with hopes to call his wife and young son back in Romania that night, he resigned himself to finish the job.

The task was one he had done many times before and involved a series of tests measuring the safety and integrity of the electrical wiring. It should have taken about an hour. Sandu says the request came to him verbally, via an intermediary. A protective suit was locked away in an equipment room. The trip would have added another 15 minutes to the job, he says, and people were waiting. In the rush, Sandu decided not to retrieve the suit.

Once in the voltage room, Sandu says the managers confirmed the lines had been de-energized. In the first round, everything went as planned. Then, on the second test, while holding insulated tongs, he linked the cables. But this time, the lines were live.

The blast was immediate and catastrophic. In milliseconds, 400 volts and 12,000 amps hit Sandu’s face and across the right side of his body. A colleague who was down below the platform and witnessed the incident told Investigate Europe that he “heard a loud bang, saw a lot of smoke, and saw [Sandu] screaming in pain.”

“I went completely blind,” Sandu recalls. “I told the managers three times that I could not see.” But no managers answered, he claims. According to the witness who called for help, untangling Sandu from his harness in a mass of blackened flesh, the managers had run off, while Sandu writhed in pain on the ground.

Workers rushed to carry Sandu to an ambulance. When doctors received him in the Grand Hôpital de Charleroi, he was placed in an induced coma for five days.

Hospital records seen by Investigate Europe show that the arc flash electrocution had burned 12 percent of Sandu’s body, with second-degree burns to the face and neck as well as third-degree burns across his right shoulder and fingers. Photographs show scorched ears and swollen raw lips, with his arm a mess of hematoma and clotted skin.

When Sandu woke in the intensive care unit a week later after surgery, his wife was at his bedside. He says his entire body felt like a “white-hot blister…. It was so much pain. The human heart cannot comprehend it,” he recalls.

Two days after waking from his coma, the regional director of Google, along with a representative from Dornan and his Dutch recruitment agency, Wanted, came to visit. Sandu says that while he lay bandaged in the hospital bed, Google and Dornan each took a statement from him about the accident. The Google director told him he was lucky to be alive.

Sandu’s electrocution is both brutal and harrowing. Yet to date, no fines or sanctions are known to have been issued. Belgium’s federal labor inspectorate closed its investigation into the accident at the end of October and found no wrongdoing. Sandu says, however, that he was never interviewed directly by the labor authorities.

Despite repeated formal requests for his accident report from Wanted and Google, he has yet to receive it.

A month after his accident, the recruiter messaged Sandu to see how he was. They then asked: “[The company] is recruiting more testers (perhaps you know people?).”

A Google spokesperson told Investigate Europe: “Our supplier code of conduct and contracts require all suppliers to comply with local labor laws and uphold safe, healthy working conditions. We maintain active dialogue with union leaders and trade groups, and are committed to providing transparent information about our workplace."

A Wanted spokesperson said: “As an employer, we can, will and should be held accountable by the proper authorities. Although we agree that freedom of speech and press is a great asset, we cannot comment on the content besides the fact that your statements or claims are not correct.”

 with media partners in 9 countries into working conditions on data center sites across Sweden, Norway, Belgium and the Netherlands reveals that Sandu’s case is far from unique.

In extensive interviews with migrant workers at flagship sites where large US tech companies are operating, reporters found evidence of debilitating injuries, squalid living conditions, an unreported infectious outbreak, unpaid wages and exploitative employment practices.

Workers at one major Microsoft project described a culture of fear and intimidation where construction managers “rule on mental terror“. The testimonies are supported by dozens of health records, pay slips, phone messages, photographs and videos shared with reporters.

The investigation found a recurring pattern where tech companies and their many Irish construction partners embed US-style working practices, often in defiance of national working norms.

In Italy, Denmark, Finland, Sweden, Belgium, the Netherlands and Norway unions told reporters they had difficulty accessing certain sites to speak with workers and document conditions. One union representative in Denmark said some sites are “more closed than a state prison.”

The European data center industry is experiencing an infrastructure surge the continent has not seen in decades, dispatching tens of thousands of migrant workers to make its vision of tech sovereignty a reality.

A data center construction site can contain up to 2,000 workers and dozens of contractors, with up to a quarter of a million people working on sites across Europe every day.

The industry is one of big budgets and even bigger secrecy. Construction firms are often required to keep their tech clients secret, while contractors and intermediaries are also typically hidden from view.

At the same time, construction continues to rank as the most dangerous economic sector in the European Union. According to 

, the sector recorded 776 deaths in 2024, representing nearly a quarter of all fatal workplace accidents in Europe.

the industry is forecasting €100 billion of investments

 into European data centers. With over 2,300 facilities rapidly altering the physical and energy landscape of Europe, 

and hundreds more planned or in construction

, the industry’s priorities appear abundantly clear: where to build, how fast to build, and which companies are prepared to build under budget.

“I wake up at two in the morning, and I can hear rats eating, moving in the walls. This house is full of rats. When I put pizza in the oven, they are running in the walls near me,” says Adrian*, a 45-year-old construction worker on a Microsoft data center site near Gävle in Sweden.

A video Adrian shared with Investigate Europe shows a large rat scurrying across a hallway and into an electrical storage room. In a string of text messages sent in August, Adrian begs the housing manager to get rid of them. The reply comes back: “Brother, I understand. But I have to wait for these other companies to come in and remove them, I can’t come in alone and just pull them out of the walls…”

If you head northwest out of Gävle, a town two hours from Stockholm, past timber-framed cottages and dense pine forests, a sign points you to a smooth road rolled out specifically for Microsoft. The remote clearing is part of 

a 130-hectare, 33.7 billion Swedish kronor (€3 billion) expansion plan

 started in June 2024 to scale up Microsoft’s AI data center infrastructure in the country.

Signs for Collen Construction line the drive, their banners draped across the large metal perimeter. Collen is among many Irish contractors overseeing data center build-outs across Europe. Hired by big tech clients to deliver projects on a strict timeline, these years-long projects manage a syndicate of other firms.

In the distance, beyond the security fencing, an airport-sized building looms, reflectionless and inscrutable. In the foreground, a bright red turnstile entryway welcomes workers in white lettering: ‘Thank You for Working Safely’.

One of Collen’s subcontractors, the Irish-owned construction company Benbau, employs hundreds of workers across Europe, including in Sweden, to work in areas such as site management, concrete and carpentry.

The workers are housed 20 minutes off-site in modular housing managed by Benbau.

For months, there has been a growing sense of outrage among Adrian and his colleagues working for Benbau on the Microsoft site. They have reported vermin infestations, broken ventilation systems, and lack of heating during the sub-zero Swedish winters, but they claim the problems remain unresolved.

One 50-year-old Romanian worker, who is diabetic, shared images of his toes that had lost circulation after wearing what he claims were poor work boots issued by Benbau. He had to have his toe amputated several months ago.

Adrian says it has been months since he first reported the rat infestation, and the rodents have multiplied inside the 40-room modular housing unit. Another video appears to take the viewer down a narrow hallway into a shared kitchen and lounge, where surveillance cameras are mounted in the corners. Inside the crowded housing, workers say the stress runs high: 24 people to a kitchen, 14 men to a single bathroom, with a boarded-up hole where the rats crawled through and a bathroom window that is sealed shut.

Such crowded confines can make the transmission of disease more likely. In February this year, one Romanian worker on the site became extremely ill. Following several trips to the hospital, it was confirmed he had active tuberculosis, Investigate Europe was shown the individual’s medical record confirming the positive diagnosis. Shortly after, others showed similar symptoms.

After the first worker who contracted the disease had left for hospital, workers in the accommodation became worried about possibly also getting the disease.

One worker who had recently moved into a new room said: “I bought bedding myself, washed the mattress with chlorine, scrubbed the walls, left the window open for a long time - it was very cold.”

Benbau initially informed only a few men who shared close quarters with the first infected worker and then told the men testing was optional, according to seven workers who spoke with Investigate Europe. Freedom of information requests to Sweden’s national work authorities and Public Health Agency revealed no outbreak had been reported.

The county medical officer in Gävle told Investigate Europe that only one case was reported in March and shared with the national health authority, which also confirmed the case. In Sweden, tuberculosis is a 

 that is subject to mandatory contact tracing, which workers claim did not occur.

Eight Benbau workers living in the same housing and union representatives told reporters that they believed at least 10 to 12 men also had the infectious disease.

Mihai*, a 60-year-old worker, also tested positive for tuberculosis, hospital records shared with Investigate Europe reveal. He says that Benbau offered to transfer him to another Microsoft site in Denmark, along with others he claims were also infected. Considering his age, Mihai says he decided against traveling.

Then, in May 2026, witnesses who lived in the housing where some were confirmed positive with latent tuberculosis say a group of vans arrived to relocate workers to the Denmark site. Two of the witnesses say that many of those who left that day had tested positive for tuberculosis. Some men were diagnosed with latent tuberculosis, yet told reporters they experienced symptoms weeks before.

Others who left that day, including an entire crew who worked in concrete for Benbau, had tuberculosis symptoms, the witnesses claim. In total, approximately 40 men left for Denmark that day, according to witnesses.

The Danish construction union 3F Byggegruppen says their records show around 45 posted workers arriving at the Microsoft site in Høje Taastrup around this time. No further record of what happened to any of the men is available, despite efforts from 3F to reach Benbau employees.

The 50-year-old worker who moved into the bedroom says that he had to get tested in Romania. He said that he was scared that if bosses found out he was speaking to the media, “that I would be sent to Denmark too.”

Workers say for the first few years, the working environment at Collen and Benbau wasn’t too difficult. But in the last two years, with the accelerated build-up of AI infrastructure, the mood has shifted. “It now feels like we are in the army service,” one Romanian worker told Investigate Europe in a video call with other colleagues. “They rule on terror, mental terror.”

Workers allege they are subject to a regime of daily harassment and regularly shifting rules. Many say the language barrier creates ongoing problems. To monitor workplace safety, the general contractor, Collen, introduced a football-style violation system. These orders, implemented by Benbau, consist of ‘yellow cards’ and ‘red cards’ for perceived infractions.

‘Carded’ workers describe being interrogated by managers for hours, then made to write letters admitting wrongdoing to workplace incidents even if they feel they have not done anything wrong.

“We are forced to sit in a room and confess to things we had not done,” one worker claims. Following a violation, they are then sent to attend a ‘reinduction’ training in English, and not their native Romanian.

Under another program, called ‘Goodsaves’, workers are required to report site or worker issues twice a week. Failure to do so results in threats of further pay deductions and extra training, texts seen by Investigate Europe show.

“These are control mechanisms,” insists one worker who asked not to be named.

“On Friday there’s the first message: ‘You have by 12 o’clock to finish your [reports]’. Then after 12 the threats begin: ‘You need to do your [report], do you want to work on Saturday?’ Then the last threat is: ‘We are going to cut the [missing report] out of your wages.’”

A former manager from Romania who worked for Benbau, who had decent housing and a wife and young child on site, told Investigate Europe he “refused to participate in the interrogation sessions” as a translator. He believes the written confessions were “designed to humiliate” other colleagues.

“In that atmosphere, I couldn’t continue seeing people working like this,“ he says. “I did not want to be afraid of someone who is forcing men to work 13 hours a day. I talked with my wife and I said: ‘This is the moment we need to go back to our country.’”

European unionization encounters US Big Tech

The living conditions and working climate form only part of the story. 

Over 30 workers shared hundreds of pay slips with Investigate Europe, many show them working up to 60-hour weeks with no overtime pay, in an apparent violation of Swedish labor law. Some workers claim they have received no paid holiday as required under Swedish law for five years.

Solidariska byggare (Workers in Solidarity), a union group, has tried to recover years of backpay for migrant workers at the Gävle site, who have been paid up to 30 per cent less than their Swedish counterparts.

Historically, Sweden has operated under a self-regulated labor market system where trade unions and employer organizations set wages and working conditions through collective agreements rather than state-mandated laws. Unions now say that over the years this model is beginning to not only erode, but crack wide open.

As data center sites expand, the ever-lengthening chain of subcontractors involved becomes more challenging to track.

Documents obtained from a freedom of information request to Sweden’s Work Environment Authority reveal 148 firms working under Collen at the Gävle site, including 28 direct subcontractors. One subcontractor can have more than a dozen firms beneath them.

In Sweden, it is customary, though not legally required, for clients to demand a collective agreement clause in their public procurement calls for tender. Yet when Collen broke ground in 2019 for the Microsoft Gävle site, no such agreement was in place. Soon after, many of its subcontractors, including Benbau, also passed over collective agreements with unions.

With the accelerated influx of US multinationals building in Europe, many companies working under them are willing to forgo collective agreements entirely, according to unions. With foreign workers unfamiliar with the laws, and the contractors setting the terms, inequity can arise quickly.

“It is a very clear picture,” says Pelle Sunvisson, a union leader at Solidariksa byggare. He describes what he sees when handing out leaflets outside the Microsoft site in Gävle.

“At 4 o’clock there is a wave of Swedish workers. They are typically electricians working with Swedish companies that have Swedish collective agreements. If you wait an hour, then there is a smaller group of white-collar Irish workers leaving, and then around six you get some Polish, Romanian, Serbian workers. Then at seven or eight o’clock, all the workers [from outside Europe].”

Over the past several months, dozens of Romanian workers at Benbau have begun joining Solidariskans byggare. Negotiations with Benbau to obtain unpaid wages recently stalled, and the union considered holding a blockade at the site, accusing the company of withholding nearly 10 million Swedish kronor (€900,000) in wages and back pay for 78 men.

“Since migrant workers don’t know the system, there is a big black hole in the Swedish model, where it becomes like a lawless territory,” says Sunvisson. 

“We have people cheated, who don’t get their salaries, even the low salary that has been promised. Some people [in the construction industry] are working five months for €500, with physical and verbal violence.”

Data obtained from freedom of information requests to the Swedish Work Environment Authority show that since 2019, the Irish company Winthrop has been fined 12 times for failing to meet core reporting requirements for foreign workers at its Swedish data center sites. In one unannounced inspection in 2021, authorities noted 15 different nationalities working at Winthrop on the Microsoft site in Sandviken.

Since 2021, the Work Environment Authority has opened 50 cases into data center sites involving Microsoft and Irish firms Winthrop, Collen and Mercury. The records, dating from 2021 to 2026, show a number of risks identified, including a serious danger of workers being struck by passing vehicles and machines falling into excavations.

Other inspections noted a lack of fixed fall protection, including suitable guardrails and platforms. One inspection of Collen in 2024 resulted in an immediate stop-work order and a conditional fine for failing to take adequate measures around fall protection. It stated the company’s lack of protection posed an immediate and serious danger of personal injury.

In October 2021, a forklift ran over the foot of a supervisor working for Collen in Gävle. The worker subsequently needed surgery. An incident report submitted by Collen to work authorities found the wrong machine had been used, the driver had no formal certificate for it and the lift plan was incomplete. Still, it had been signed off. In its accident report, the subcontractor listed both the injured man and the driver as self-employed.

The Work Environment Authority saw no grounds to report the employer to prosecutors.

When approached for comment, Collen declined to address specific issues raised but a spokesperson said the allegations were “wholly inconsistent with Collen’s values, policies and the standards to which we hold ourselves and those with whom we work.”

“We are committed to ensuring that every Collen site is a safe, healthy and respectful place to work, where everyone is treated with dignity and their fundamental human rights are protected,” the spokesperson added. “We expect the same throughout our supply chain and have zero tolerance for any conduct inconsistent with these principles.”

‘Beyond this gate, it is not the Netherlands’

It is not just in Sweden that workers have reported problems. Pyotr Zaitsev*, an engineering manager, worked on data center sites for another Irish construction firm in Germany until 2019.

Pyotr was highly experienced, and his daily tasks involved managing electrical installations while supervising groups of subcontractors. But he would regularly work beyond his contracted hours, he claims. He had “a contract for 40 hours a week on paper,” but in reality, “I was actually working 60 hours minimum, up to 80 hours from time to time.”

Hearing that the working conditions were better in the Netherlands, he found a job at a number of large Winthrop sites there.

Yet after 18 months, the pressure at Winthrop began to match that of his former employer, he claims, saying he had “severe burnout.” One day in 2021, after working a full shift, Pyotr says he fainted on site from overexertion. In an email to Winthrop’s HR department sent the following week, he informed the company he had gone to the hospital without specifying why.

Winthrop said they had no record of such an incident. "It is company policy that all adverse events must be reported, investigated and comprehensively recorded.”

He also said he had growing concerns for workplace safety. There was once a serious incident with a Romanian subcontractor, he claims: “There was a [mobile elevated working platform],” he describes, “the [the worker] was distracted and he was using the remote control and he drove over his [own] leg.”

Winthrop said they conducted an analysis and that the employee returned to work the next day, adding that it adhered to all applicable laws. “Winthrop is committed to maintaining the highest standards of health and safety across their projects, ensuring our people, partners and clients can operate in a safe and supportive environment,” a spokesperson said. “We place safety at the heart of project delivery.”

Pyotr has been on sick leave from the company since 2023. In May 2025, a Dutch government employment body ordered Winthrop to give Pyotr extended sick pay until May 2026, after finding the company had not fulfilled their obligations in helping him return to work. Winthrop stopped paying him in September 2025.

A Winthrop spokesperson said: “Despite repeated efforts to re-integrate the employee since he left in 2023 following a non-work-related accident outside the workplace, no progress was made in terms of the employee resuming work.”

Accessing data center sites to inspect worker conditions can be challenging, according to FNV, the Netherlands’ largest trade union.

Its representatives claim they attempted to access the Winthrop site in Schiphol-Rijk in September to speak to workers. At the gate, a Winthrop employee refused them entry, reportedly telling them, “Beyond this gate, it’s not the Netherlands… I am not allowed to let you in by upper management.”

FNV claims it has made several unsuccessful attempts, including an email request in summer 2026, to visit the Winthrop-Microsoft site in the past year.

A spokesperson for Winthrop said that union representatives were granted access through “pre-arranged protocols” across their data center sites.

“Any suggestion that unions are denied access to sites is false and misleading,” the spokesperson said.

The vast majority of Dutch construction companies are bound by a collective labor agreement that allows unions access to the sites. Foreign companies are not automatically covered by those agreements.

‘Sites are more closed than a state prison’

Belgian, Swedish, Danish, Finnish and Italian unions all told this investigation that they have had trouble accessing data center sites or have been shut out entirely.

Lotta Sjölander, a representative at the national trade union Byggnads in Sweden, the largest in the country, says representatives have been unable to access the Microsoft sites in Gävle and nearby Sandviken “from the very beginning.”

The Danish construction union 3F Byggegruppen, comprising 56,000 members, is proud of its success in getting worker parity. Yet Christian Foli Jensen, a 3F union consultant

 says gaining access to sites is a growing challenge. The problem, he says, is that “the clients don’t follow the normal way to do things, like on construction sites in Denmark. The sites are more closed than a state prison.”

Roberta Braga is founder and executive director of the 

Digital Democracy Institute of the Americas

For years, the debate over AI regulation in the United States has been framed as a false dichotomy. Some argue if we move too slowly on innovation, the US risks falling “behind” its competitors. Others say if we regulate technology too aggressively, innovation suffers.

New polling of 1,500 US Latino registered

 voters by the Digital Democracy Institute of the Americas (DDIA) suggests Americans don't see those goals as mutually exclusive.

In particular, the highly consequential Latino electorate that tends to swing between the parties is looking for something more practical. Latinos want progress, but they also want rules that protect their jobs, personal information, and communities. This is the case for Latino Republicans, Independents, and Democrats.

Where are Latinos on AI and data centers?

Latino voters are adopting AI, led by young men between the ages of 18 and 29, but familiarity with the technology has not brought comfort. More Latino voters are worried than excited, and most expect AI to shape their lives.

82% of respondents believe that what happens with AI in the coming years will affect what happens in their life. Latino voters’ concerns about AI are mostly connected to perceptions about impacts on everyday life. 74% of respondents worry that AI will take jobs away from ordinary people, 72% worry about their personal data when they use AI, and 71% say it’s hard to tell what is and isn’t made by AI on the Internet. Fears about AI causing human extinction or a global catastrophe in the next few decades are less common, though still shared by 60% of people.

When it comes to data centers, our research found that 74% of Latino voters worry data centers will drive up electricity bills. Another 68% believe they put too much strain on local water supplies. By comparison, only 40% believe data centers bring good jobs and investment to the communities where they are located. Just 42% agree that building more data centers is necessary for the US to stay ahead of China and other countries in AI.

The gap between perceived costs and benefits is striking and becomes even clearer when the question turns local. Just 23% support building a data center in their own community, while 51% oppose it.

There are important partisan differences. Latino Republicans are more likely than Democrats to believe data centers bring jobs and investment, 55% to 32%, and that they are essential to US competitiveness, 58% to 32%. Yet even among Republicans, 72% want stricter rules governing data center construction and 64% worry about higher electricity bills.

This is where AI policy has an opportunity to move beyond the familiar innovation-versus-regulation divide. A serious infrastructure governance agenda should consider enforceable requirements for assessing electricity and water impacts, meaningful public participation in siting decisions, greater transparency around promised economic benefits, and safeguards against shifting infrastructure costs onto households.

It should also ask whether communities have the ability to negotiate the terms under which AI infrastructure is built, rather than simply being informed after the most consequential decisions have been made.

What, specifically, do Latinos want to see from policy?

One of the clearest findings from our research is that Latino voters overwhelmingly want stronger oversight of AI, regardless of political affiliation. Nearly eight in ten respondents (78%) believe there is a pressing need for stricter rules and oversight of AI. This includes 83% of Democrats, 73% of Independents, and 72% of Republicans.

Support extends to specific proposals. Two-thirds favor requiring companies to submit risk assessments to the US Department of Commerce before releasing new AI models. An equal share supports independent review and approval of AI systems before release. And 61% support pausing the development of more advanced AI until appropriate safety standards are established.

These findings challenge the idea that public support for technological progress requires a light regulatory touch. They also raise questions about the reliance on voluntary commitments and industry-led safety frameworks.

When asked whom they trust to keep people safe as AI develops, only 19% of Latino voters express trust in technology CEOs, and just 22% trust AI companies. Scientists are the only group tested that more respondents trust than distrust, with 50% expressing trust. Government institutions fare poorly, too. Only 27% trust Congress, and 28% trust the president to keep them safe as AI develops.

These very low levels of trust in elites and institutions create a difficult governance challenge. The public wants stronger oversight, but has little confidence in either the companies developing AI or the institutions responsible for regulating it. Trust must be earned through transparent, enforceable safeguards, independent scrutiny, and mechanisms that allow the public to hold both companies and regulators accountable.

What can policymakers take from these findings?

Days from the US midterms and as a future Congress prepares to take office, our poll findings provide a useful starting point for examining AI and online safety policy that serves Latino communities in a low trust environment.

First, policymakers should advocate for requiring meaningful risk assessments before powerful AI systems are released.

 67% of Latino voters support having companies submit assessments to the US Department of Commerce before releasing new models, for example. Such assessments could examine how models enable deceptive content, impersonation, privacy violations, and risks to minors, including their potential use in misleading civic information.

Second, policymakers should consider requiring that AI companies have their systems reviewed by independent entities rather than relying exclusively on companies to evaluate themselves.

 67% of Latinos we polled also support this approach. Reviews could assess the accuracy and reliability of AI systems across languages, including the widely spoken Spanish, and evaluate whether safeguards work in practice.

Third, policymakers should push to further improve transparency around AI-generated content.

 75% of Latinos say it’s hard to tell what’s true and what’s false on social media these days and 71% agree it is hard to tell what’s AI-generated online. People want more tools to help them navigate pollution online. Practical standards for provenance, labeling, and source attribution could help people better understand where information comes from. Such measures should acknowledge that labels and watermarks have limitations and cannot independently establish whether a claim is true.

Fourth, policymakers must place children's safety at the center of AI product development.

 83% of Latinos polled worry about the safety of children online. Age-appropriate design, stronger privacy protections, and independent testing of risks to younger users deserve consideration before products are widely deployed.

Finally, there is a broader lesson in these findings that point toward a different model of public engagement. Rather than consulting communities only after regulations are drafted or infrastructure decisions are nearly finalized, policymakers should incorporate public perspectives earlier and more systematically. That means investing in accessible public consultations, including Spanish-language engagement, supporting independent research into how different populations experience AI, and ensuring that community concerns have a meaningful pathway into regulatory decisions.

It also means recognizing that expertise is not confined to technology companies, academic laboratories, or government agencies. Workers, parents, educators, journalists, and local residents have knowledge about how these technologies operate in everyday life that should inform how they are governed.

The US is approaching a consequential period for AI governance. Per our data, neither political party in the US owns the issue of AI and data centers. Latino voters’ views on AI lean toward caution and regulation, positions where Democrats currently have an edge. But with about one in five saying they trust both parties nearly equally, and about one in eight saying they trust neither party, there is still an opening for either party to own the issue with this community.

As investment accelerates and the technology becomes more deeply embedded in everyday life, policymakers must make decisions about safety, infrastructure, economic competitiveness, and the distribution of benefits.

The temptation will be to frame these decisions as a choice between moving fast and falling behind. Our research suggests that this is the wrong choice to offer the public.

The next generation of AI governance should be judged by whether it establishes enforceable accountability, protects communities from foreseeable harms, and gives the public a meaningful role in shaping the technology's development.

A common assumption in AI governance is that greater transparency regarding the composition of the datasets used to train large language models will enable governments, researchers, and other stakeholders to demand greater accountability from the companies that build AI systems. Training datasets are often a 

 of AI related harms, such as bias, discrimination, and the erosion of digital rights. The 

 of transparency requirements is to provide copyright holders, privacy advocates and critical AI researchers information about the data used to train AI models, so that they can verify whether their interests were protected, and hold AI companies to account. In doing so, dataset transparency disclosures can arm consumers (both individuals and companies) with the necessary information to choose which AI system is right for them.

This perspective has been adopted by multiple state, national and international jurisdictions. Most notably, 

 have both passed laws requiring frontier AI companies to provide summaries of their training dataset composition. But many stakeholders focused on governing AI in the public interest may not sufficiently understand or appreciate the limits of current training dataset transparency mandates. This piece aims to demystify what existing dataset transparency mandates actually produce, detail critical processes they fail to reveal, and then point towards productive avenues for pursuing meaningful transparency about datasets that can better enable accountability.

It is important to understand that uncovering the contents of training datasets alone is insufficient to enable meaningful accountability. Knowing 

 is included in a training dataset provides little insight into 

 it was included—let alone how it can be contested. And the rationales behind data practices ultimately shape whose data is included, whose interests are protected, and who bears the burden of dataset-related harms. Achieving accountability requires understanding 

 these artifacts were made, and the rationale behind dataset creation practices that are often considered banal or perfunctory.

Existing training dataset transparency statements are vague and ineffective

California's Training Data Transparency Act (TDTA), which came into effect Jan. 1, 2026, mandates that developers of generative AI systems provide a “

” of training dataset composition across twelve categories, from the types of data included to the presence of private data or copyrighted material, how the data was acquired, any cleaning, filtering or manipulation processes, and the presence of synthetic data. The European Union's 

 into effect, also includes a requirement that developers of generative AI systems publish a “sufficiently detailed summary” of the contents of their training datasets. However, compliance with these mandates has been far from straightforward. 

 against AI developers has incentivized opacity about training dataset creation and contents. In March, a California federal court 

 xAI’s legal challenge of the TDTA. The company argues the law forces it to reveal trade secrets in violation of the Fifth Amendment and compels speech in violation of the First Amendment—a decision that 

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe