Surveillance capitalism in 2026. The concrete infrastructure
Dear Reader,
In 2019 the Harvard business scholar Shoshana Zuboff published a book called The Age of Surveillance Capitalism. It was, and remains, the definitive account of what happened to the internet economy in the 2010s. Zuboff coined the term surveillance capitalism to name the specific business model that had emerged, in which the raw material of the economy was behavioral data extracted from users, and the product sold was predictions about what those users would do next.
At the time the book was published, the description was already accurate. Seven years later it is no longer just accurate. It is understated. The infrastructure Zuboff described has been built out, refined, and integrated across sectors in ways that the 2019 version of the book anticipated but could not yet document. And the specific mechanisms have gotten more effective, more subtle, and more difficult to resist.
I want to write about what surveillance capitalism actually looks like in 2026, in concrete terms rather than abstract ones. Because the abstract discussion has been going on for years and has not produced much action. The concrete discussion, I think, is more useful for anyone trying to figure out what is being done to them and what to do about it.
Start with the core mechanism, because it is worth stating precisely.
Surveillance capitalism is a business model with three components. First, extract behavioral data from users through their interactions with digital products. Second, use that data to build predictive models of what specific users will do in specific situations. Third, sell those predictions to third parties who want to influence user behavior, or use them directly to influence user behavior yourself.
The middle step, the predictive modeling, is what distinguishes surveillance capitalism from earlier forms of advertising. Traditional advertising was untargeted. It broadcast messages to whoever happened to be watching. Direct marketing added targeting based on demographic data, but the targeting was crude. What surveillance capitalism added was targeting at the level of individual predicted behavior, powered by the specific data that only comprehensive digital surveillance could produce.
This model is now dominant in the digital economy. It funds most of the free services people use every day. It funds a substantial portion of paid services too, because even paid services often collect data that gets sold or used for the same predictive purposes. The specific practices vary, but the underlying business model is the same across the sector. Behavioral data is the resource. Prediction is the product. Behavioral influence is the outcome.
What is important to understand is that this is not a byproduct of the digital economy. It is the digital economy. The specific reason companies collect so much data, track so many behaviors, and build such elaborate profiles is that behavioral prediction is where the money is. Everything else is downstream.
Let me walk through what the specific infrastructure looks like in 2026, sector by sector.
Advertising is where the model started, and it is still the largest single application. When you see an ad on a website or in an app, you are seeing the output of an auction that occurred in the milliseconds before the page loaded. Advertisers bid on the right to show you an ad based on the specific predictive profile the platform has built about you. Your interests, your recent behavior, your predicted willingness to buy, your predicted response to specific messaging, all of this goes into the auction. The winning bid gets to shape what you see next.
This infrastructure has become more sophisticated over the last five years. Real-time bidding has been supplemented by longer-horizon predictive modeling that considers not just what ad you might respond to right now, but what sequence of ads over weeks or months might shift your buying behavior. The specific technique is called causal advertising, and it is being deployed at scale by the largest ad platforms. Not just to influence individual decisions but to shape longer arcs of behavior.
Social media is the second major application. What you see in your feed is not chronological, and has not been for over a decade. It is the output of a predictive model that estimates what content will keep you engaged. This model uses your entire history of interactions, plus data from users who look similar to you, plus data about how content of different types performs. The feed is optimized to hold your attention, which means to maximize the platform's revenue. This is why the specific content you see is often more emotionally intense than what your friends actually post. Emotional intensity holds attention.
E-commerce is the third. Product recommendations, dynamic pricing, personalized offers, and search result ordering are all outputs of surveillance capitalism infrastructure. When you shop online, the specific products you see, the specific prices you are offered, and the specific promotions you are shown are all tailored to you based on predictions about your behavior. Two users looking at the same page can see meaningfully different prices, and often do.
News and media are the fourth. Every major news outlet uses reader behavior data to shape editorial decisions. Which stories get promoted, which headlines get tested, which specific articles get emailed to which subscribers, all of this is now data-driven. This is not just about audience research. It is about optimizing revenue by maximizing engagement, which shapes what news you actually see.
Finance is the fifth. Credit scoring, insurance pricing, loan approval, and investment recommendations are increasingly driven by behavioral data. Not just traditional financial data but data about how you use the internet, what devices you own, who your contacts are, what you post on social media. This data feeds into predictive models that make consequential decisions about your access to money and services.
Politics is the sixth. Political campaigns now routinely use behavioral targeting at scale, based on the same underlying infrastructure that advertisers use. Voter modeling, targeted persuasion, and message testing are all versions of the same surveillance capitalism approach applied to political outcomes.
Employment is the seventh. Hiring algorithms, performance evaluation, and internal team recommendations are increasingly driven by behavioral data collected in workplace software. Every keystroke, every mouse movement, every email pattern feeds into models that shape hiring, firing, and promotion decisions.
Add all of these together, and what you have is not a sector-specific pattern. It is the operating system of the modern economy. Surveillance capitalism is not something that happens in advertising. It is something that happens everywhere the economy touches digital systems, which is now nearly everywhere the economy operates.
There is a specific structural fact about this system that most people miss.
The primary customers of surveillance capitalism are not the users. Users are the product. The primary customers are the third parties who want to influence user behavior. Advertisers. Political campaigns. Employers. Financial institutions. Anyone who wants to shape what users will do, at scale, with high precision, in ways users may not notice.
This is not a metaphor. It is the specific business relationship. When Facebook makes money, the money comes from advertisers, not from users. When Google makes money, the same is true. When most digital platforms make money, the money comes from whoever wants access to the users, not from the users themselves. The users pay in data. The paying customers pay in currency. The platform intermediates.
Which means that everything the platform does is oriented toward serving the paying customers, not the users. The specific ways user experience gets shaped, the specific features that get built, the specific outputs that get optimized, are all downstream of what will produce the most value for the actual customers. Users get whatever incidental benefits the platform can deliver while still serving its actual customers. Sometimes those benefits are substantial. Sometimes they are minimal. The relationship is not symmetric.
This structural fact explains a lot of user experience. Why the platform seems to be doing things that are not in your interest. Why features you like get removed. Why the specific quality of the platform seems to deteriorate over time in specific ways. All of this is downstream of the fact that you are not the customer. You are the raw material.
There is a specific development in 2026 that Zuboff could not have fully documented in 2019, and that has become central to how surveillance capitalism actually works.
AI systems, particularly large language models, have become deeply integrated into the surveillance capitalism infrastructure. Not primarily as user-facing products, though there are increasing numbers of those. Primarily as the analytical engines that process behavioral data and produce predictions.
Traditional behavioral prediction used statistical models trained on structured data. These worked, but they were limited to the specific data types they were trained on and the specific patterns that could be extracted from that data. AI systems can process much more heterogeneous data. Text that a user has written. Images they have shared. Audio from voice interactions. Video from cameras. Sensor data from devices. All of this can now be processed by systems that extract semantic meaning and integrate it into unified user profiles.
The specific consequence is that the predictive infrastructure is now much more capable. What it knows about you is not just what you have clicked on. It is what you have said, what you have shown interest in, what you have implied about yourself in the specific ways you communicate. And the predictions that come out of this infrastructure are correspondingly more precise.
This has happened relatively quickly. Five years ago, most behavioral prediction was still relatively crude. Today, the best systems can predict specific individual behaviors with accuracy that would have been considered science fiction ten years ago. The specific arc of improvement is roughly the same as the arc of improvement in AI capabilities generally, because the two are the same phenomenon applied to slightly different problems.
For users, this means that the influence being applied is more effective. The ads that hit are more likely to convert. The recommendations that appear are more likely to be acted on. The messages that arrive are more likely to shift behavior in the intended direction. Whatever defenses individuals had against earlier surveillance capitalism are less effective against the current version.
Let me tell you about a specific thing I noticed recently that made this concrete for me.
I bought a book about a topic I had not previously written about, not searched for, and not discussed online in any way. The book was recommended to me by a specialist on an unrelated project, and I bought it on a specific website in a private browsing session.
Within three days, I started seeing content related to the book's topic across multiple platforms. Instagram ads for related products. Recommendation feeds that included articles on the same subject. Email newsletters I subscribed to that suddenly featured pieces on the topic. Nothing about my purchase should have been visible to any of these systems. And yet, somehow, they all knew.
The specific mechanism was probably that the bookseller had shared purchase data with one or more data brokers, and the data brokers had provided that data to the platforms I use. This is legal. It is standard. It happens with essentially every online purchase, and increasingly with offline purchases too, through loyalty programs and payment systems.
What was disturbing was not the fact that it happened. I know how these systems work. What was disturbing was the specific texture of being tracked at this level. Something I had done in what I thought was privacy had been observed, correlated, distributed, and used to shape what I would see afterward. And this was for a mundane purchase, not something sensitive. The same infrastructure processes everything, all the time, without any signal that it is happening.
I share this because it made the abstract discussion concrete for me. The infrastructure is not hypothetical. It is not something that will happen if we are not careful. It is happening. It is happening to me. It is happening to you. And most of the time it is happening without either of us noticing, because the specific effects are calibrated to be too small to detect at any single moment.
There is a specific implication for AI that I want to name.
The infrastructure of surveillance capitalism is now the substrate on which AI is being deployed. Almost every consumer-facing AI application is trained on data that was collected through the surveillance capitalism pipeline. The predictive models that AI systems produce are used within the same pipeline. And the specific optimization objectives that AI systems are trained against are usually derived from surveillance capitalism metrics.
Which means that when we talk about AI ethics, AI safety, and AI alignment, we are talking about systems that are already embedded in an ethical and economic structure that has its own dynamics. AI is not a neutral technology that will be shaped by our choices about how to use it. It is already being shaped by the specific economic structure into which it is being deployed. And that structure has its own logic, which is not going to yield to abstract ethical considerations.
This is why some of the more thoughtful voices in AI ethics have been arguing that the AI alignment problem cannot be separated from broader questions about the political economy of technology. The specific challenges of aligning AI systems with human values are inseparable from the specific challenges of building an economic system that would make such alignment possible. If the economics of AI development require extraction of behavioral data at scale, then AI systems will be optimized for extraction, whatever their designers say they intend.
This is not a call to abandon AI development. It is a call to notice that the technical work of alignment is not sufficient on its own. Without corresponding changes in the economic structure, the technical work will be undermined by the incentives it operates within. Both need to change. And the political will to make the economic changes has, so far, been almost entirely absent.
What can be done about surveillance capitalism at the individual level.
Less than most guides suggest, but more than nothing.
You can use tools that limit tracking. Privacy-focused browsers, tracker-blocking extensions, VPNs, encrypted communications. All of these help at the margin. None of them provide comprehensive protection, because the infrastructure is designed to be difficult to escape.
You can reduce your engagement with the platforms that extract the most data. Fewer social media accounts. Fewer apps. More offline activity. This helps too, though it also isolates you from social and professional networks that assume you are participating.
You can use paid services when possible. Services you pay for have less reason to extract data from you, though they often extract it anyway. Still, paying is generally better than free from a privacy perspective.
You can support policies that would regulate the surveillance capitalism infrastructure. This is the only intervention that would actually change the system at scale. And it has been very slow to develop, because the economic interests aligned against regulation are enormous.
I do most of these things. I am under no illusion that they provide meaningful protection. The infrastructure I am describing knows more about me than I know about myself, and it has been building that knowledge for over a decade. What I can do is limit further extraction and support systemic changes. What I cannot do is undo what has already been collected and modeled.
I want to close with what I have come to think of as the specific ethical failure of the current moment.
We have built an infrastructure that watches everyone all the time, predicts what everyone will do, and shapes what everyone sees in order to influence what everyone does. We have done this without asking whether people wanted it, without seriously considering the consequences, and without building any institutions capable of governing what we have built.
The specific reason we have done this is that it makes money. The economic engine that drives the infrastructure is powerful enough that the questions about whether it should exist have not been effectively raised. When they have been raised, they have been dismissed with reference to consumer benefits, or with claims that regulation would harm innovation, or with the observation that users seem to keep using the platforms despite complaining about them.
None of these responses actually engage with the core question, which is what kind of society we want to live in. A society where every action is observed, every prediction is monetized, and every user experience is optimized to shape behavior is a specific kind of society. It is not the society most of us thought we were building. It is not the society most of us would choose if we were choosing directly.
But we are not choosing directly. We are participating in an economic system that produces this outcome mechanically, without any single decision to produce it, and without any real opportunity to say no. This is the specific way that democratic and market institutions have failed to keep up with technological change. And it is why the question of what to do is not just an individual question but a collective one, requiring collective institutions we do not currently have.
Next month I want to write about doomscrolling specifically, because it is the specific behavioral pattern that most exemplifies what surveillance capitalism produces at the level of individual experience. Understanding doomscrolling as a system-level outcome rather than a personal failing is essential to responding to it. Stay with me.
— Transmission Sent —
Niklas Hanitsch
Reference materials
- Shoshana Zuboff — The Age of Surveillance Capitalism (2019)
- Bruce Schneier — Data and Goliath (2015)
- Cathy O'Neil — Weapons of Math Destruction (2016)
- Kate Crawford — Atlas of AI (2021)
- Frank Pasquale — The Black Box Society (2015)
- Chris Wylie — Mindfck: Cambridge Analytica and the Plot to Break America* (2019)
- https://www.hbs.edu/faculty/Pages/item.aspx?num=56791
- https://foundation.mozilla.org/en/privacynotincluded/
Continue reading
- Dopamine loops: How social media hacked human motivation
- The attention economy is eating consciousness
- Six ways AI is already manipulating your decisions
- The algorithmic god: How recommendation systems became our post-religion
- AI Surveillance: The Observer Watching the Observer
Frequently asked questions
What is surveillance capitalism? Surveillance capitalism is the business model in which companies extract behavioral data from users, build predictive models of user behavior, and sell those predictions or use them directly to influence user behavior. Shoshana Zuboff coined the term in 2019, and it has since become the standard framework for analyzing the modern digital economy.
Are users the product or the customer? Under surveillance capitalism, users are the product. The customers are third parties who pay to access users through advertising, targeting, and predictive services. This structural fact explains much of user experience under these systems, including why platform features often seem misaligned with user interests.
How has surveillance capitalism changed since 2019? The infrastructure has become more sophisticated, particularly through the integration of AI systems that can process much more heterogeneous data than earlier statistical models. Predictions are more accurate. Behavioral influence is more effective. The reach has extended from advertising and social media into finance, employment, politics, and most sectors of the economy.
Can I opt out of surveillance capitalism? Not really. You can limit your exposure through privacy tools, reduced platform engagement, and paid services, but the infrastructure is now pervasive enough that comprehensive opt-out is essentially impossible for most people. Systemic change through regulation is the only intervention that would substantially reduce the practice.
What is the connection between AI and surveillance capitalism? AI systems, particularly large language models, have become deeply integrated into the surveillance capitalism infrastructure as the analytical engines that process behavioral data and produce predictions. This means that AI development is embedded in an economic structure whose incentives shape what AI systems get built and how they are optimized. AI ethics cannot be separated from broader questions about the political economy of technology.