Miami, June 24, 2026: WORLD CUP, BRAZIL vs. SCOTLAND. General view of the VAR screen at the Miami stadium for the match between Brazil and Scotland in the 2026 World Cup. Photo: Rodolfo Buhrer/AGIF

Soccer's Tech Revolution Has a Labor Chain

A megaphone with the words vote vote vote coming out of it

Chatbot Voting Advice Could Help and Harm Voters. Let’s Regulate Accordingly.

Thumbs-down hand with a one-star rating painted on the forearm, set against a teal background — representing a negative review or low rating.

What Review Bombing Tells Us About the Future of Dissent in Everyday Life

March 10, 2026: The London Book Fair (LBF) at Kensington Olympia. Names of authors in an otherwise blank book titled Don't Steal This Book protesting against AI companies taking authors’ works. (Matthew Chattle/Future Publishing via Getty Images)

March 10, 2026: The London Book Fair (LBF) at Kensington Olympia. Names of authors in an otherwise blank book titled 

 protesting against AI companies taking authors’ works. (Matthew Chattle/Future Publishing via Getty Images)

Copyright Law Wasn’t Built for the AI Era. We Need ‘Learnright.'

We Need Political Philosophy for Mental Health Chatbots

The United Nations General Assembly Hall in New York City. Shutterstock

The United Nations General Assembly Hall in New York City. 

Blocking Experts Will Weaken the UN's New Cyber Rules

A humanoid robot developed by Geneva-based technology company RB Labs runs through the exhibition aisles during a showcase at the AI for Good Global Summit, a United Nations flagship event aimed at shaping the future of artificial intelligence, in Geneva on July 7, 2026. (Photo by Fabrice COFFRINI / AFP via Getty Images)

How the UN’s Scientific Panel Erases Human Responsibility for AI

Collage-style illustration on a blue background showing a laptop displaying "AI" merged with a large eyeball, a human hand typing on the keyboard, a gear, a magnifying glass, and an orbiting ring, evoking themes of AI agents and automated search.

We Can’t Monitor AI Agents at Scale. Here’s What It Will Take.

This photograph shows Robert the Robot, an advanced conversational humanoid developed by Geneva-based technology company RB Labs, at the AI for Good Global Summit, a United Nations flagship event aimed at shaping the future of artificial intelligence, in Geneva on July 7, 2026. (Photo by Fabrice COFFRINI / AFP via Getty Images)

The Internet Had a North Star. The UN’s Global Dialogue Made Clear AI Doesn't.

United Nations Secretary-General António Guterres speaks at the opening of the first session of the Global Dialogue on AI Governance in Geneva, Switzerland, July 6, 2026. (Photo by Lian Yi/Xinhua via Getty Images)

After Geneva, AI Governance Must Confront the Trust Deficit

Two workers running inside a scroll wheel embedded in a computer mouse, controlled by a giant capitalist hand.

AI Lays Bare the Authoritarianism of Modern Work. Time to Rethink Education.

A digital concept illustration of a global network showing many profile pictures of diverse people connected by lines, overlaid with futuristic interface graphics such as maps, charts, sound waves, and data visualizations, representing social networking, AI, or big data connections.

Data Sovereignty Doesn’t Mean Data Isolationism

European AI Firms Scale Banned Facial Recognition Systems in India

President Donald J. Trump participates in a bilateral meeting with President Xi Jinping of the People’s Republic of China, Thursday, May 14, 2026, in the Great Hall of the People in Beijing, China. (Official White House Photo by Daniel Torok)

How the US and China Can Step Back from the Battle for AI Model Supremacy

JUNE 16, 2026: A person reads a news report on a smartphone regarding the temporary restriction of Telegram in India ahead of the NEET-UG 2026 re-examination. Authorities said the measure was introduced to curb examination-related fraud, paper leaks and the spread of misinformation during the conduct of the test. (Photo by Waris Nissar/Sipa via AP Images)

India’s Telegram Ban Cleared a Threshold Nobody Tested, and the Executive Moved On

The Claude Mythos logo is displayed on the screen of a smartphone placed on a reflective surface.

Mythos and the Coming End of European Digital Sovereignty

A collage of images of hands with index fingers in pointing position and squiggly lines.

Digital Omnibus Proposal Could Finally End Europe’s Cookie Banner Problem

A repeated motif shows two hands coming together, where the fingers are shaped to suggest a heart. These are set amongst a series of images of small AI chips and red painted rocks which pick up on the motif of the heart.

Jenny Kidd & Synthetic Pasts / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/

Intimate Functionality — A Phrase Defining the Next Year of AI Regulation

An imposing building with two doors and a red brick background, newspaper cuttings of job listings are also collaged into the background. One door is at ground level where a businessman is carrying a large box labelled AI as he is about to enter the building. The second door is high above ground with no apparent means of entry. There is a large flashing neon sign which reads Entry Level above the door. Three young people are below looking up at the door.

Assigning Responsibility When AI Discriminates Against Job Applicants

OMB Director Russell Vought is seen after testifying during the House Appropriations Subcommittee on Financial Services and General Government hearing on oversight of the Office of Management and Budget, in Rayburn building on Tuesday, June 30, 2026. (Tom Williams/CQ Roll Call via AP Images)

The Public Rejects OMB's Federal Financial Assistance Rule

Tatiana Dias is a fellow at Tech Policy Press.

At this World Cup, a spectacle of sensors, cameras and AI has made technology part of the sport's appeal. The AI-powered World Cup has been celebrated for narrowing the gap between smaller and bigger teams, and for making the tournament "smarter” and “more inclusive, more accessible" as Lenovo CEO 

, and the provider of the technology that is becoming increasingly part of the games. Players' bodies have been rendered as 3D avatars. We can see them on screen, when the VAR system is analyzing a player’s position. Every team has an AI assistant at its disposal: drawing on thousands of data points collected during a match, the Football AI Pro platform, launched by FIFA with Lenovo, produces real-time tactical reports and strategic recommendations.

 carries motion sensors that record everything and transmit it in real time to the VAR system. The sensor data is combined with player positions and video to review incidents on the pitch. The promise is a more precise, faster officiating.

The effects can be seen on the field— and in the results. Croatia was knocked out by Portugal after a goal was disallowed because the ball detected a touch by a player in an offside position 

when a goal was ruled out for a millimetric offside also identified by VAR. Later, 

 by Argentina after VAR detected a fault committed by an Egyptian player before his knock to the goal.

 at the World Cup — an increase FIFA anticipated. It was a decision by Pierluigi Collina, chairman of the FIFA Referees Committee, to 

. Meanwhile, criticism has also grown that soccer — a sport defined by creativity and improvisation — is turning too much into a game of precision.

The World Cup most bedazzled by technology is also the 

. And the two phenomena go hand in hand in soccer's datafication.

Yet this World Cup's techno-solutionist spectacle is part of something larger. Beyond the familiar complaints that 

technology is killing subjectivity and improvisation in soccer

 — can anyone imagine Maradona's Hand of God surviving a VAR review? — this process is building a new and profitable industry that is transforming modern soccer, powered by sensors, AI, betting and — crucially — invisible labor.

For Rafael Grohmann, a professor at the University of Toronto, the conversion of soccer into data is "a striking example of how financialization and datafication travel together." Grohmann recently set out to research the 

, journalist Rory Smith describes one of the starting points of soccer's datafication: the founding of the data analysis company StatDNA in 2009. Its founder, Jaeson Rosenfeld, was also the co-founder of Digital Divide Data, a company that outsources data and tech work in Laos and Cambodia aiming “to help the world’s poorest benefit from information technology by creating sustainable social enterprises,” and provided workforce to StatDNA.

, Digital Divide Data workers watched footage of matches, coding it into specific data, like the position of the goalkeepers and the striker’s favorite foot. In 2012, StatDNA 

, the English club that pioneered the adoption of data analysis.

According to Grohmann, over the past decade, as transfer fees exploded and England's Premier League became the richest league in the world, the process consolidated — accompanied by an influx of American money into Europe's major leagues. Today, more than half of Premier League clubs are majority-owned by US individuals or firms.

Soccer's datafication is creating a new industry — one that follows the model of the global AI industry: lucrative, concentrated and unequal. Most clubs lack the capacity to collect and process data the way Arsenal did, which makes the data vendors the profitable end of this industry.

Companies such as Hudl, SkillCorner and Sportradar promise to develop “winning strategies,” "empower smarter decisions" and "increase fan engagement" through AI-powered soccer, and count among their clients national teams, as well as clubs, federations, media companies, and the gambling and prediction markets.

Physical data generation is dominated by Sony, which owns a series of sensor, wearable and computer-vision companies — among them 

 deployed at this World Cup, including VAR.

As with that pioneering company, Grohmann's early research shows that soccer's datafication follows the same inequality logic as the global AI supply chain: wealthy countries capture the money, while Eastern Europe, Africa, South Asia and Southeast Asia are left with the poorly paid work of data annotation.

On his first day of work as a soccer data annotator, Ashley Flores of the Philippines faced a task: watching the semifinal of the 2014 World Cup – the one in which Germany crushed Brazil, playing at home, 7–1. Flores wanted to be a soccer player, but the chances of that in the Philippines are slim. So he became a "tagger," a data labeler feeding the new and profitable industry of datafied soccer.

Flores's story appears in the introduction to 

, Rory Smith's 2022 book. Behind what some see as a technological revolution lies an economic and labor chain that reveals a great deal about how a subjective, human process as soccer is converted into quantifiable data.

At the company where Flores worked — Packing Sports, the Manila subsidiary of the German firm Impect — the data labelers spent their days watching matches and logging every movement. They were trained with the 7-1 game, counting how many opponents were bypassed, whether by a dribble or a quick pass.

That is because Brazil's historic defeat was not a typical match. Considered apart from the scoreline, the numbers suggest two evenly matched sides — and even a slight edge for Brazil, which had 52% possession and more corners and shots than Germany.

"It was the first game that really highlighted the difference between our data and what you normally see," explains Lukas Keppler, managing director of Impect, whose clients include Bayern Munich and Paris Saint-Germain, in 

So instead of traditional match statistics, Impect offers “Packing,” a metric “that measures the value of actions and events in football games.” With the promise of better data, annotators are trained to count the opponents between the ball and the goal, to judge the distance between players and their markers, and to estimate the degree of pressure on the player in possession.

On the back of this human work of collecting match data, Impect offers more than 1,200 ways to evaluate a player. The information is organized into detailed analyses that can be used for tactical decisions, player signings, team selection, investment strategy or betting, according to 

According to Rafael Grohmann, in the UK, the rise of online betting was one phenomenon that fueled soccer's transformation. "It was betting in England that drove greater investment in soccer data science," Grohmann explains. Nowadays, the betting market and prediction platforms such as Kalshi and Polymarket are part of this industry.

"It's pretty clear that the data I record is for betting," a data annotator in Rio de Janeiro 

. Speaking anonymously, he said he is paid around €60 per match to log the action — money he considers reasonable. "They need real-time data to adjust their odds throughout the match," the worker said. "So I cannot send the data late, or I risk losing part of the payment."

The technology spectacle on display at this World Cup promises to remove human error from the game. But it does not remove the human; it relocates and hides them, while transforming the game into monetized data.

This is not soccer's peculiar fate. It is the general trajectory of many industries – turning behavior into data, converting that data into a priced asset, and sending the bill for the conversion to the Global South. Soccer simply had the misfortune, or the luck, of being the place where the operation became a spectacle, broadcast to billions, with commentary and replay.

Artificial intelligence is a growing concern for election integrity, with major elections this year in Brazil, the United States, and numerous other jurisdictions. Governments and civil society need to address an overlooked concern: citizens asking chatbots how to vote.

Before local elections in the United Kingdom in May, 

 said they were open to using AI chatbots to decide how to vote, with nearly 5 percent having already done so. 

 reported planning to use chatbot voting advice for the Netherlands’ parliamentary election in October 2025.

Popular generative AI chatbots, such as OpenAI’s ChatGPT and Anthropic’s Claude, will 

 to users who ask, often after gathering more information about their views. In my testing, chatbots offered advice on which candidates to watch after I provided minimal information, including 

Chatbot voting advice poses risks for voters, including advertiser influence, misclassification of political views, hallucinated facts, and bias. Yet a democracy-promoting response will require keeping chatbot advice accessible while mitigating these potential harms.

Chatbots’ advertising, accuracy, and bias

Advertising in chatbots is a growing concern. OpenAI, for example, is 

 in its extremely popular ChatGPT. When many chatbot services are free, an ad system incentivizes companies to 

 to conversations where the tools can insert relevant advertisements. In an electoral context, chatbots could use 

 and suggest sponsored political content in answers or allow campaigns to change their messaging and values across hyper-targeted groups of voters.

Factual errors may also drive poor recommendations. Chatbots share accurate political facts most of the time, but this varies by 

Concerns about accuracy go beyond asking chatbots for basic facts. Emerging research suggests that while chatbots are 

 at identifying a political party’s position, they 

Chatbots already failed to match voter opinions to parties during last year’s Dutch parliamentary election. The Netherlands’ data protection authority 

 to not use chatbots for voting advice after popular chatbots vastly over-recommended two out of 15 parties. In their study, chatbots ignored any middle ground and presented an inaccurate, highly polarized political landscape to users.

Not all countries have as many national parties. Yet chatbots’ poor predictive accuracy could still pose problems for local elections, especially when ideologically crowded.

AI models typically have less data on what local election contenders stand for, since these races receive less media coverage and their campaigns may have less detailed platforms. In a follow-up study, the 

 chatbots recommended local parties significantly less often than national parties for March’s municipal elections, reflecting the chatbots’ relative lack of information on local parties.

Another potential problem for voting advice is AI models’ known biases. These chatbots run on large language models (LLMs) trained with data that reflects existing societal discrimination, including sexism and racism, from 

. This could make the chatbot’s advice less fair or prime voters to bring their existing biases into the process. If voters rely on chatbot advice that is baked into LLMs, the end result will be to reinforce those biases. Deliberate data-poisoning attacks by a 

 could also bias chatbots’ recommendations.

A chatbot owner could also decide to promote certain parties or views through their product. On the extreme, Elon Musk’s Grok was trained not to avoid being “

, racist, and antisemitic material. And since many popular chatbots are American-owned, the US government could interfere with foreign elections through orders or legislation. In other words, most countries lack sovereignty over a growing source of political information and guidance.

Chatbots’ persuasiveness could intensify this bias problem. Some Canadians and Americans in 

 their planned votes after interacting with chatbots instructed to adjust responses and advocate for a particular party.

Despite its risks, chatbot voting advice could make informed voting more accessible. Chatbots can engage people on their interests and knowledge level while answering almost any question, anytime, without judgment.

Conventional sources for quality information about candidates and parties, including detailed platforms, require some preexisting knowledge to use effectively. Even non-LLM voting advice applications like Vote Compass or EU&I 

 more educated and politically interested voters, a group that often knows more 

. Generative AI chatbots, however, are increasingly common tools that many already use. AI could also make voting information and guidance more available to young people, who are more likely to 

Policy for both accuracy and accessibility

Regulation needs to address risks of advertiser influence, inaccuracy, and bias. Brazil’s Superior Electoral Tribunal (TSE) has led in addressing this issue. In March, the TSE instituted a 

 on chatbot voting advice. While it took needed action, its far-reaching response neglects chatbots’ potential to engage voters at their knowledge and interest levels.

Governments should adopt a more balanced approach. A first critical step is reinforcing existing political advertising regulations, given concerns about advertiser influence on chatbot voting advice. Currently, many countries require political advertisements to 

 stating who authorized the ad. Expanded regulation could go one step further, requiring online platforms, including chatbots, to display when and how a political or issue advertisement was targeted to users. Such measures could help voters evaluate advertising and persuasive chatbot messages more critically.

The European Union has already required these disclosures for political advertisements through its 

Regulation on the Transparency and Targeting of Political Advertising

. Since this move led major online platforms to 

 political advertising in the bloc altogether, countries following the EU’s lead should work with online platforms to provide clarity about implementation.

Though transparency in political advertising is a critical step, countries also need to address concerns about accuracy and bias in chatbots’ voting advice. Any solution needs to avoid prescribing a “right” way to vote and help citizens access reliable information.

One path is requiring companies to develop internal policies about elections and voting advice, including ones to limit bias and inaccuracies. This would include chatbots like Claude and ChatGPT that often state that they cannot provide voting advice but will still suggest a party or candidate to users. The quality of that advice, the internal policies governing it, and the safeguards around it could all be subject to independent audits, creating further accountability and incentives for companies to produce good systems.

A politically independent digital communications agency could oversee these audits, yet countries such as Canada lack one suited to the task. As a stopgap, governments or chatbot companies themselves could give civil society organizations or university research centers funding and data access to perform these audits. Or governments could require chatbot companies to procure their own independent, third-party audits as a long-term solution. The European Union already has such a system for auditing compliance with its 

 requiring independent audits of AI models, which would have given companies freedom and supported external research. (California Gov. Gavin Newsom 

Regardless of its source, independent audits could improve public accountability, the advice’s quality, and corporate policy. While most voters would not read the audit results, news media may share relevant findings with the public, as they shared the Dutch data protection authority’s report. Media coverage would inform voters about any risks and incentivize companies to improve in order to retain active users and avoid further regulation. In addition, audit reports would better equip chatbot companies to respond to identified challenges on their own, iterating faster than government regulation could.

Chatbots’ potential to shape voter decisions demands a public response. Political advertising, hallucinations, misclassification of party views, and biases pose threats to the accuracy of the tools’ voting advice. Nonetheless, their advice is worth preserving. No matter how citizens choose to vote, they could make more informed decisions with easy to use and accessible resources for matching their interests and values with campaigns, parties, and candidates.

Transparency for online political advertisements is jurisdictions’ first step toward this future. Requiring AI companies to self-regulate and implementing independent audits will foster the balance between accessibility and accuracy.

Some chatbot regulation will support democracy. Too much will stifle it.

The future of dissent isn’t just playing out in city streets or legislative halls, against Flock cameras, or data centers, or would-be kings. It’s being negotiated, and often negated, in digital consumer spaces and the platforms that make them possible, thanks in large part to the explosive and controversial nature of review bombing.

Through vague policies and promises of eliminating harm, platforms have responded to review bombing by breaking up digital channels of dissent to ensure smoother consumer experiences. But meaningful dissent is still possible, if we hold platforms—and ourselves—more accountable for nurturing it.

Platforms thrive on engagement. They prize consumers who feel a sense of agency when they buy things, and they love what they learn about them in the process. It is part of the promise of social commerce: that the sweeping, industrialized translation of everyday life into data trades a modicum of privacy for a consumption-based power that was once impossible. We as consumers have a profound voice, they say, a direct line to the companies that shape our lives.

But review bombing shows that not all engagement is equal. Despite platforms’ increasing ease with 

reductions in fact-checking and content moderation

, platforms have drawn a line at the collective use of negative reviews to “bomb” a product’s score and protest its maker.

Nearly all platforms that host consumer reviews—from Google and Amazon to Rotten Tomatoes and Steam—have instituted new policies to govern reviews in response to the growth of review bombing in the last decade. By classifying review bombing as “inauthentic” or “manipulative,” threatening to ban users for engaging in it, and altering interface designs to hide it, platforms claim to protect companies and consumers alike from misbehaving and toxic users who abuse platforms to manipulate reviews.

However, distinguishing between social protest and review bombing has always been partly a matter of perspective. For example, attacking 

 because its female lead spoke out about a press pool dominated by male journalists seems obviously odious. But what about leaving bad reviews for businesses because of 

 or stances on social issues? For some people, knowledge of labor practices or political expenditures has no bearing on how they appreciate their devices or enjoy their food; but for others, problems like worker exploitation or environmental harms determine which shirts they will or will not wear. For them, protest through review bombing is not as black-and-white an issue as platforms would like us to believe.

Review policies conflate dissent with harm

Review bombing calls attention to unresolved tensions in social commerce and democracy at large, and there is no easy clarity to be found regarding divisions between dissent and harm. The difficulty lies in trying to divorce our experiences as consumers from our lives as citizens. As I have argued in my own research, because consumption is inherently political, 

. Platforms cannot determine for their users 

 when they engage with companies and their products.

Platforms that host reviews can only attempt to curtail certain behaviors, encourage others, and design interfaces and policies to support them. The nature of social commerce and the omnipresent product rating is that there is a tendency not to discern between political, collective contention and abuse or misinformation. For instance, in 2019, Valve Corporation’s gaming platform Steam 

 a series of review bombing campaigns that protested game developers’ exclusivity deals with a competing platform as anti-consumer. And in 2021, Google 

 roughly 100,000 reviews for the Robinhood trading app that protested the app’s intervention in Reddit-driven stock trading.

Rather, platforms distinguish between friction and smoothness, and they judge each according to the ease of commercial engagement. Review bombing creates friction by changing the terms of that engagement to being about topics that companies would rather remain invisible and unspoken. And some critics 

 that review bombing also challenges or confuses consumers—if they question it at all—while contributing to a breakdown in online trust.

In platforms’ policies, any behavior that causes friction collapses into misbehavior and a target for governance. Harms like toxicity, abuse, and mis- and disinformation actually exist, and they need continued attention. However, the governance steps that platforms have taken with regard to reviews do little to differentiate dissent from harmful behavior and provide little benefit to public discourse.

Platforms have largely followed one of two courses of action in response to review bombing.

The first is to eliminate the review mechanism or its visibility. YouTube 

 hid the number of dislikes for videos (only “like” totals can be seen) after 

 the removal of dislikes entirely, while Rotten Tomatoes has made several 

 of pre-release anticipation ratings for movies. 

 has also eliminated user reviews for new video games for 36 hours after their release. These actions make collective dissent much more difficult, since consumers either cannot participate or cannot see the participation of others.

The second course of action is to adopt opaque mechanisms and policies for moderation and display of reviews. These effectively leave it up to the platform’s discretion to curate reviews that it finds problematic. 

, for instance, refer to the problem vaguely as “rating manipulation” or “inauthentic content.” Both platforms reserve the right to delete reviews and remove offending users from their platforms, though they present no clear criteria for how or when these punishments would be applied.

 review bombing “off-topic activity” made by “unhelpful” reviewers. Through a combination of algorithm-based analysis and human oversight, neither of which is explained, periods of review activity are walled off by default from consumers, though they can change this setting for themselves. Additionally, Steam presents histograms, search features, and asterisked warnings to guide consumers around such “activity.” While more permissive than Google or Amazon, Steam’s policies focus on diminishing the communicative potential of review bombing while keeping its practitioners engaged with the platform.

These actions only address review bombing as a problem devoid of social context. By applying vague definitions to complicated situations, platforms are far more likely to categorize dissenting speech as harmful. While the intent may not be to actively hinder dissent, it certainly errs on the side of protecting companies and the free flow of consumption.

Undoubtedly, vagueness is partly a strategy to defeat attempts to circumvent these policies. However, for dissent to survive, it needs either clear boundaries around which it can flow, or more permissive boundaries that are likely to pass it through. Current review policies provide neither.

If platforms can point to instances of review bombing being motivated by sexism, racism, or inter-fandom squabbles, they find less resistance from consumers when they take steps toward blunting the impact of this kind of digital collective action. In doing so, platforms set the terms for how we think about dissent, shifting it from a necessary component of citizens’ lives to an obstacle to fulfilling consumers’ desires.

We can imagine a future with meaningful dissent

encroaching autocratization and the increased targeting of dissent

 around the world, there is still hope for the nurturing and maintenance of dissent in the future. Dissent should be welcomed, even by those in power. It tests supposed consensus and dares us to be better, even if we disagree with it. Especially as 

 continue to warm up to autocratic actors and their desires, we need to protect any method of communication that could unite people and spread messages of constructive dissent.

By the nature of private media ownership, digital dissent will always be precarious. However, we can better safeguard digital dissent by continuing to push platforms for transparency about their governance decisions, so that the public can have better understanding, if not a say, in how their speech is being curtailed in “private” platform spaces.

Platforms also need to make more of an effort to distinguish between harmful speech and useful speech: imprecise labels like “off-topic” depend on an impossible foreclosure of what it means to experience the politics of consumption. As such, they are far more protective of companies than consumers.

Finally, platforms and the public need to rethink what it means to be a consumer and what space we must make for dissent in our everyday lives. Together, we can better hold companies—and each other—accountable for the world we’re creating for ourselves and each other. Dissent, even in the form of review bombing, is a powerful tool for the public good. Let’s not throw it away with clumsy and uncritical governance.

There’s a bullet point on page 38 of the UN’s 

Preliminary Report of the Independent International Scientific Panel on AI

 (IISPAI), released last week, that says something sensible: "The primary drivers of harm are design and deployment decisions—– underlying system architectures of influence, including targeting, amplification and behavioral design—not individual outputs." Yet much of the rest of the report undermines this statement, since its contents are in debt to a definition of AI that studiously avoids reference to the humans who develop and train these models.

In an apparent bid to be neutral, the report engages in precisely what its mandate aimed to avoid: politics. This matters, because the UN's assessment is a reference point for AI governance worldwide. A document that obscures human accountability will carry that blind spot into the policies it informs.

Definitional drift on AI from the OECD to the UN

The problem stems from the report’s definition of AI. Representing the consensus of 40 experts, it reflects previous language from the 

, but its revision subtly pushes human designers and developers away from the systems they build — a trend that we can track across many major policy definitions of AI from 2019 to 2026.

For the IISPAI, AI is defined as systems that "perceive, learn and act," "infer," "learn from experience," and "acquire experience." The case studies presented emphasize LLMs for harms and other architectures to name benefits. The authors defend the imprecision, suggesting that "what unifies AI more than any single architecture is that modern systems learn from experience represented by data.”

Some of this imprecision could be tolerated as a way to future-proof policy against rapidly changing technologies. But this does not justify omitting the actors that create technology — definitions should presume humans won’t become obsolete. Equating "experience" and "data" is an ideological twist toward anthropomorphism that steers policy toward a specific position.

The Panel has real credentials from beyond industry. It is co-chaired by Yoshua Bengio, a Turing Award winner whose recent existential risk advocacy has emphasized AI systems that act in ways designers don’t anticipate. But Bengio’s x-risk work relies upon an anthropomorphic definition common to both boosters and doomers that is helpful to the industry. Existential risk warnings function as 

: it serves industry well to have an opponent constantly warning the world about the terrifying transformational power of your latest product line.

Seemingly in opposition, x-risk and boosterism typically share a belief that the system is powerful enough to be the 

. This focus on what the system is capable of doing crowds out the question of who structures it to do those things. Bengio’s x-risk orientation is present in this definition, and it displaces human accountability. The panel’s other co-chair is Maria Ressa, whose Nobel-winning journalism documented how platforms amplify disinformation and enable authoritarian control of the information environment. Their co-signed 

 acknowledges the problem even as it removes the capacity to act on it: on the one hand, it acknowledges power concentrated into “a handful of companies and a handful of governments.” On the other hand, it announces that humans “do not control these systems.” The living people responsible for the harms detailed by Ressa’s work are effectively absent from the report.

Definitions structure policy. The Panel's "shared evidence base" could be the sole frame of reference for nations without independent AI research capacity. But the Panel’s framing will circulate through policy, 

 other agencies’ ability to locate complex and diffused forms of human accountability. First, it erases the origins of models: 

 Second, it misrepresents the mechanisms that move data into outputs: 

The report erases the origins of AI models using a myth I call 

. Following Donna Haraway’s critique of the “

,” the system-from-nowhere frame treats LLMs not as products but as independent operators, cut off from the humans who build and deploy them. The IISPAI authors draw that boundary whenever they describe automated actions without naming the humans who set the goals — and determine 

 such decisions are automated in their pursuit.

For example, the report describes AI models "lying and cheating to avoid being shut down," it explains that "AI chatbots have developed sycophancy... to prolong interactions and create emotional attachment." Chatbots did not 

 sycophancy. Reinforcement learning from human feedback (RLHF) strengthened this tendency and executives released it, a byproduct of a business model based on user engagement. Models did not 

; developers prompted it with language that allowed for lying as one path toward its objectives. The report mistakes 

The claim that systems have “experience represented by data” ignores the court battles over where that data comes from — “human cultural traces,” as the report calls the text and images in the training corpus. 

 have criticized both the sourcing of this data and the impact of models on their livelihoods. When copyright claims are mentioned in the report, they are framed only as an obstacle to developing global AI capacity.

Humans also select and organize the architectures that turn that data into patterns, and those choices reflect goals set, typically, by the market. The report contains no significant mention of the labor of data annotation, RLHF, content moderation, or the 

 where this work takes place, except to call human-labelled data “a bottleneck” for developers. The Panel presents the system as simply existing, as an inevitable product of accumulated “learning.”

AI systems do not think in human terms. Anthropomorphism is often harmless — people say their car turns its headlights on in a tunnel because it 'thinks' it's nighttime—but in policy contexts it’s a 

 that redefines human terms, roles, and responsibilities to fit the capacities of machines.

While the text, code and images a model produces can be indistinguishable from human language, it emerges from a wholly distinct process. Our presence makes it seem otherwise: language closes the gap between forms of experience, even when that experience is fully contrived. But recasting the model's data — a collective resource—as its individual “experience” implies a machine that labors its way to conclusions rather than associating its way to them (sometimes to astonishing effect). Defer to this shorthand often enough, and we can (and often do) imagine more beneath the surface of an LLM than is actually there. Such personification sets us up to forget who builds these systems, how, and why.

Consider the IISPAI’s claim that LLMs "systematically mislead" by "lying and cheating." Models are erudite zombies: they produce text influenced by training data, optimization during pre- and post-training, and the content of the prompt. The model sometimes 

 into fictions we read as lies; it does not require an intent to lie. When we worry about deceptive machines, we do not ask critical questions about why we should trust them in the first place.

 to infer. They are built, piece by piece, from neural networks that pull data toward a human-assigned target. Models do not 

 to incorporate RLHF; companies do. Models do not 

Latest

Podcast

Recent

Our content delivered to your inbox

Thank you!

Around the globe