AI Didn’t Create Surveillance. It Completed It.

Privacy is dead.

Not because every private conversation is being watched at this exact moment, nor because one government possesses a single perfect database containing the complete life of every person. Privacy is dead because the conditions that once made it possible have disappeared. Human life now produces a continuous digital record, that record can be retained almost indefinitely, and artificial intelligence has removed the final practical barrier that protected us: the inability to understand all the information being collected.

Surveillance used to be expensive. It required time, people and attention. A government could follow a political dissident, investigate a suspected criminal or monitor a foreign agent, but it could not follow everyone. A police officer had to watch a building. An intelligence analyst had to read an intercepted message. A clerk had to locate a file. An informant had to remember what had been said and write it down. Photographs had to be recognized by human eyes. Records held by one institution might never be connected with records held by another.

Even the most intrusive governments faced a physical problem. They could collect enormous quantities of information, but they could not read everything, remember everything or connect everything. Their ambitions exceeded the capacity of their bureaucracies.

That limitation is gone.

Artificial intelligence did not invent state surveillance. It completed the machinery required to make mass surveillance work.

The internet solved the collection problem. Smartphones, cloud computing, digital payments, social networks, cameras, advertising systems, connected vehicles and commercial data brokers transformed ordinary human behavior into an expanding archive. AI solves the analysis problem. Language models, facial recognition, graph analytics, computer vision and automated data integration can turn scattered records into identities, relationships, timelines, predictions and operational decisions.

We have spent the first years of the generative-AI revolution arguing about whether a chatbot can write an essay, imitate an artist or eliminate an office job. These are visible consequences, and some are important. But they are not the largest use of the technology. The most consequential use of AI may be its ability to convert accumulated information into power.

The Bottleneck Was Never Collection

Edward Snowden’s disclosures in 2013 revealed that the United States had built surveillance systems on a scale few citizens understood. The Privacy and Civil Liberties Oversight Board subsequently described an NSA program that had collected millions of domestic telephone records in bulk. Another surveillance program collected electronic communications when targets were believed to be non-Americans located outside the United States. Snowden did not reveal the beginning of mass surveillance. He revealed that its infrastructure was already operating.

The public debate that followed concentrated understandably on collection. How much information should intelligence agencies be permitted to acquire? What constituted a search? When was a warrant required? Could metadata expose a person’s associations even when the contents of a conversation remained unread? Was the collection lawful, necessary or effective?

Those questions remain essential, but they concealed a second problem. Collecting information does not automatically produce knowledge.

A billion telephone records do not interpret themselves. A warehouse full of intercepted messages is not an intelligence assessment. A decade of location data is not yet a biography. Somebody—or something—must identify the relevant people, resolve conflicting identities, translate different languages, connect transactions, reconstruct sequences, determine relationships and decide what matters.

For most of the history of surveillance, that work was performed by human beings. Computers made searching faster, but human attention remained the scarce resource. Intelligence agencies could collect more information than their analysts could reasonably examine. The data became larger while the number of hours in a human life remained unchanged.

AI was built for precisely this imbalance.

The NSA publicly states that its research includes data science, machine learning and scalable analytic techniques designed to extract actionable intelligence from raw data. Its work spans human-language technology, computer vision, large-scale graph analytics and collaboration between analysts and AI agents. This is not a critic imagining what intelligence agencies might eventually do. It is the agency describing what it is developing.

The Office of the Director of National Intelligence has instructed the eighteen agencies of the American Intelligence Community to make their data interoperable, discoverable and ready for use by both people and machines. Its data strategy explicitly identifies artificial intelligence as a means of managing, finding and exploiting information more effectively. Its open-source intelligence strategy says the intelligence community is already pioneering the use of AI, machine learning and human-language technologies.

The language is administrative, but the transformation is political. Once different collections become interoperable, an analyst no longer has to know where every relevant record is stored. Once information becomes machine-readable, the analyst no longer has to examine each item personally. Once language models and retrieval systems can summarize records, identify entities and answer questions across databases, the volume of collected information becomes less of an obstacle.

Surveillance becomes scalable.

Your Life Is Already Producing the File

The modern surveillance system does not require a government officer to begin following you. Your life is already generating the record.

A smartphone can record or infer location, movement, communications, browsing, photographs, financial activity, travel, application usage and proximity to other devices. Connected cars generate location and driving data. Retailers record purchases. Banks retain transactions. Advertising networks follow users across applications and websites. Cameras record faces and license plates. Cloud platforms preserve photographs, messages, contacts and documents. Health applications record steps, sleep, heart rate, exercise and sometimes reproductive information.

Most of these systems were not introduced as instruments of political control. They were introduced as conveniences.

That does not make the resulting data less revealing.

Your phone may know where you were on a Tuesday afternoon ten years ago when you no longer have the faintest idea. A transaction record may show what you bought. A search history may show what frightened or fascinated you. A map history may reveal which clinic, church, political meeting, hotel, lawyer’s office or private home you visited. Your photographs may establish who was present. Wireless and Bluetooth records may reveal which devices repeatedly appeared near yours.

The extraordinary fact is not that one device contains every answer. It is that enough fragments exist to reconstruct the answer.

The United States Supreme Court recognized part of this reality in Carpenter v. United States. The Court held that government acquisition of historical cell-site location records constituted a Fourth Amendment search. It observed that carrying a cellphone is effectively unavoidable in modern life and that long-term location information can provide a comprehensive account of a person’s movements. Such records can expose not merely where someone traveled but their familial, political, professional, religious and intimate associations.

Carpenter now has a more direct descendant. In Chatrie v. United States, decided June 29, 2026, the Supreme Court held that police conducted a Fourth Amendment search when they used a geofence warrant to obtain Google Location History data. A geofence warrant can compel a technology company to identify devices recorded within a defined place and time, beginning with a crime scene rather than a named suspect. The Court concluded that people have a reasonable expectation of privacy in their cellphones’ location records even when officers seek only a limited period and obtain the data from a third party. It did not decide that every geofence warrant is unconstitutional, or even whether the warrant used in Chatrie was reasonable. Questions about probable cause, particularity at each stage of the process and the good-faith exception were left for the Fourth Circuit on remand. The legal progression is important. Carpenter addressed a government demand for one person’s historical cell-site records. Chatrie addressed a method that searches location data first and identifies people afterward. In both cases, constitutional doctrine arrived after the technical capability had already been built and used.

The Court was responding to location records produced by cellular networks. The larger data environment has since become even more detailed. Location is now generated by applications, advertising identifiers, connected vehicles, wearable devices, Wi-Fi networks and commercial tracking services. A person may believe they are sharing information with a weather application, a navigation service or a retailer. In reality, they are helping produce a dataset capable of describing their behavior over time.

The Federal Trade Commission has repeatedly brought cases against companies accused of selling sensitive location information. In one case, the agency alleged that a data broker sold precise data capable of tracking visits to medical and reproductive-health clinics, places of worship and domestic-abuse shelters. In another, the FTC alleged that location data could identify private homes. It later acted against companies accused of selling data associated with military sites, churches, health clinics and labor organizations.

These enforcement actions establish something larger than the conduct of individual companies. They demonstrate that detailed human movement has become a wholesale commodity.

Privacy has entered the supply chain.

First-Generation Data

The easiest data to recognize are the records we directly produce.

A search.

A purchase.

A text message.

A photograph.

A location coordinate.

A telephone call.

A medical appointment.

A bank transfer.

A social-media reaction.

A border crossing.

These are first-generation data: records of observable acts.

They matter because they establish context. A person’s location at a particular time may be innocuous by itself. Combined with a purchase, a search and the presence of another device, it begins to describe an event. Add years of repeated behavior and it begins to describe a life.

This first generation of data is already more powerful than human memory. People forget. Databases do not forget unless they are made to forget. Human recollection changes, decays and reconstructs itself. Digital records preserve timestamps, coordinates, counterparties and identifiers.

The file can therefore become more authoritative than the person.

Ask someone where they were on an unremarkable afternoon in February a decade ago and they will probably be unable to answer. Ask the right collection of databases and the answer may already be waiting: the device was at this coordinate, connected to this network, moved along this route, made this purchase and appeared near these other devices.

The person no longer knows.

The system may.

Second-Generation Data

The second generation is not what you disclosed. It is what can be inferred from what you disclosed.

Repeated overnight location can indicate where you live. Regular daytime location can indicate where you work. Visits to a particular building can suggest medical treatment, religious practice, political activity, addiction treatment, pregnancy, legal trouble or a private relationship. Changes in purchases may indicate financial stress. Altered travel patterns may indicate a new job, separation, illness or caregiving responsibility.

None of these conclusions must be explicitly stored in the original data.

They can be generated.

This distinction is fundamental. Traditional privacy discussions often assume that the danger lies in exposing information a person consciously provided. AI makes that model obsolete. Institutions can infer information a person never knowingly disclosed and may not even know about themselves.

Second-generation data are created by combining first-generation records with context.

The inference may be correct. It may be partially correct. It may be completely wrong. But once created, it can influence how a person is treated.

A model does not need certainty to alter a decision. It needs only a score, a category or a recommendation accepted by someone with authority.

Third-Generation Data

The third generation concerns prediction.

What will this person do next?

Will they repay a loan?

Will they commit fraud?

Will they leave a job?

Will they attend a protest?

Will they cross a border?

Will they purchase a product?

Will they become ill?

Will they support a political movement?

Will they pose a threat?

Third-generation data do not describe an event that has already happened. They describe a possible future calculated from the first two generations. Past behavior becomes inferred identity, and inferred identity becomes anticipated conduct.

This is where surveillance ceases to be merely observational.

It becomes preventative, anticipatory and disciplinary.

A person can be subjected to scrutiny not because of what they have done but because of what a system predicts they might do. The prediction may be described as decision support rather than judgment, but the practical effect can be the same if an official treats the model’s output as authoritative.

Everything becomes data. Data become description. Description becomes prediction. Prediction becomes action.

Knowledge becomes power in its most literal form.

The Government Does Not Need to Collect Everything Itself

One of the most dangerous misunderstandings about modern surveillance is the belief that the state must personally collect every record it wishes to use.

It does not.

Private companies have spent decades constructing a commercial surveillance infrastructure for advertising, analytics, fraud prevention, customer profiling and market research. Data brokers acquire, aggregate and sell information generated across this economy. Governments can access some commercial information through purchases, contracts or other lawful mechanisms rather than building the original collection system themselves.

The Intelligence Community openly acknowledges the importance of what it calls commercially available information. A declassified ODNI report stated that the volume and sensitivity of such information had expanded because of digital technologies, including location tracking. The report recognized that commercially available information can reveal intimate details about individuals and may be combined or analyzed in ways that create powerful intelligence. ODNI subsequently issued a framework governing intelligence agencies’ access to, collection of and processing of this material because of the privacy and civil-liberties issues it creates.

The existence of a framework is itself an admission of scale. Intelligence agencies do not create specialized rules for information that is irrelevant to their mission.

Commercial surveillance and state surveillance are not identical. They operate under different authorities and pursue different objectives. But they increasingly inhabit the same data environment. Private companies collect information because it has economic value. Governments use information because it has operational value. AI increases both.

This is how privacy became wholesale. Human behavior is collected in bulk, traded in bulk and analyzed in bulk.

The state does not need to place a tracking device on every car when commercial systems already record movement. It does not need an officer to identify every face when searchable image collections contain billions of photographs. It does not need to ask every person about their associations when location, communication and transaction records can help reconstruct them.

Federal agencies are already using these systems. The Government Accountability Office reported that seven federal law-enforcement agencies used commercial or nonprofit facial-recognition services capable of searching billions of images. All seven initially used those services without requiring personnel to receive facial-recognition training. GAO also found that eighteen of twenty-four large federal agencies reported using facial-recognition technology during fiscal year 2020.

A later GAO review found that Homeland Security law-enforcement components used more than twenty detection and monitoring technologies, including facial recognition, automated license-plate readers, drones and fixed cameras. The report warned that agency policies did not always include important protections, such as limiting collection to information relevant to an authorized purpose.

This is happening now.

Palantir and the Architecture of Connection

Palantir is not the creator of this world, but it represents the type of company built for it.

Its importance lies not primarily in collecting data. Its platforms are designed to integrate information that already exists, make it searchable, map relationships and help institutions act upon it. Palantir openly markets artificial intelligence and large language models to military and defense organizations. Its materials describe software supporting intelligence operations, battlefield awareness, targeting workflows, logistics, cameras, sensors and rapid decisions across multiple operational domains.

A 2024 partnership announcement between Palantir and Anduril described a system intended to move national-security data from battlefield sensors at the tactical edge into cloud-based AI infrastructure. Palantir’s defense work is therefore not an abstract experiment involving conversational bots. It is part of an effort to connect sensing, data, analysis and action.

Palantir is especially visible because it describes much of this work publicly and has become closely associated with defense, intelligence and government data systems. It should not be treated as the sole cause. The same architecture appears throughout modern technology: collect data, unify it, identify entities, map relationships, generate insight and accelerate decisions.

The brand names will differ. The institutional purpose will differ. The machinery will remain recognizable.

What matters is the conversion of fragmented observations into operational knowledge.

Snowden Exposed Yesterday’s Machine

Snowden warned the public that the infrastructure of mass collection had been built. His disclosures forced governments, courts and citizens to confront capabilities that had developed largely outside ordinary democratic debate.

But the systems Snowden revealed belonged to an earlier technical era.

The data existed, but much of it still had to be queried through comparatively rigid systems. Analysts needed to know what they were seeking. Databases remained divided. Language, format and classification barriers made integration difficult. Human beings still performed much of the interpretive work.

AI changes the value of everything collected before it.

An archive that was too large to examine in 2013 may not remain too large in 2026. A database gathered for one purpose may acquire new significance when combined with other records. Old communications can be summarized. Historical locations can be linked. Photographs can be searched with improved recognition systems. Seemingly meaningless metadata can be reconsidered in the light of later events.

Data do not expire merely because the technology originally available could not fully exploit them.

This is one reason retention matters so much. The collection of information creates a future option. Institutions may not know how to use every record at the moment it is acquired. They may acquire that ability later.

Snowden showed us the warehouse.

AI turns on the lights.

The Stasi’s Impossible Dream

The East German Stasi constructed one of the most extensive systems of domestic surveillance in modern history. It recruited informants, intercepted communications, opened mail, recorded conversations, photographed targets and assembled immense archives.

Its ambition was to make society visible to the state.

Yet the Stasi remained constrained by paper and human labor. Officers had to recruit informants, produce reports, listen to recordings, develop photographs, index files and manually connect one record to another. The system was oppressive, but it was not omniscient. Its desire to know exceeded its ability to process.

Imagine the same institutional ambition equipped with smartphones, cloud storage, facial recognition, automated transcription, location histories, graph databases and language models.

The comparison is not a claim that every government using AI is East Germany. It is a comparison of capability. It asks what happens when technologies developed in commercial democracies are applied by a government that decides its survival is threatened.

A threatened state will use the powers available to it. It may change the law, reinterpret the law, invoke an emergency, purchase information indirectly, classify the program or insist that exceptional circumstances require exceptional measures. Different governments will use different language. The underlying incentive is the same: if information can reduce uncertainty and preserve power, the pressure to obtain and use it will be immense.

The political character of the government determines who is targeted and why.

The technology determines what is possible.

The Courts Are the Last Barrier

The most important defenses remaining to an ordinary person are institutional: courts, due process, enforceable rights, independent judges, legal discovery, public-interest litigation, journalists and organizations capable of challenging the state.

For those with money, access to those defenses is considerably stronger. They can hire lawyers, commission experts, appeal decisions and force institutions to explain themselves. The poor, the undocumented, the politically isolated and those already classified as suspicious may never discover what information was used against them.

This is why secrecy and automated inference are such a dangerous combination. A person cannot challenge evidence they are not permitted to see. They cannot correct a profile they do not know exists. They cannot cross-examine a model. They may not know whether an adverse decision came from a human judgment, a commercial database, an algorithmic score or an association inferred from someone else’s records.

The Supreme Court’s decision in Carpenter placed a constitutional limit on government acquisition of historical cell-site records by holding that obtaining those records was a Fourth Amendment search generally requiring a warrant. Chatrie carried that reasoning into the use of a geofence warrant to obtain Google Location History data: that access was also a search. But the Court did not categorically prohibit geofence warrants or decide whether the warrant used there satisfied probable cause and particularity. It vacated the judgment and sent those questions, along with the renewed good-faith issue, back to the Fourth Circuit. The legal progression is narrower than a ban and more important than a procedural adjustment. It recognizes that constitutional protection can attach not only when the government follows a known person through stored location records, but also when it searches a provider’s location database to discover who was present. The law is still trying to apply enduring protections to investigative methods that technology made possible first.

The courts matter because technology will not restrain itself. Companies will continue to monetize information. Intelligence agencies will continue to seek strategic advantage. Police will continue to seek faster investigative tools. Political leaders will continue to demand protection from enemies, disorder and perceived threats.

The question is whether any institution can still say no.

AI Completed Surveillance

AI did not create the desire to watch populations.

It did not create secret police, intelligence agencies, advertising networks, informants, dossiers or political repression. It did not invent the census, the passport, the fingerprint, the wiretap, the camera or the database.

It connected their descendants.

It removed the excuse that there was simply too much information to understand.

The modern surveillance system can collect first-generation records of what we do, generate second-generation inferences about who we are and produce third-generation predictions about what we may do next. It can search memories we no longer possess. It can reconstruct associations we never declared. It can turn commercially generated information into government intelligence and intelligence into action.

The system is not all-seeing. It does not have to be.

It only has to know enough about anyone once those in power decide to look.

For centuries, privacy survived partly because human beings could not record everything and institutions could not process everything. That was never a principled safeguard. It was a technical limitation mistaken for a permanent feature of freedom.

The limitation has ended.

The warehouse exists. The profiles exist. The commercial market exists. The government programs exist. The analytical machinery exists.

Privacy is dead because forgetting is dead, obscurity is dying and the cost of understanding an entire human life is collapsing.

AI did not create surveillance.

It completed it.

This is Part I of The End of Privacy, an ongoing NOMOTO MEDIA Investigation into artificial intelligence, surveillance, data collection, predictive profiling, and the institutions attempting to govern them.

Sources

Supreme Court of the United States, Chatrie v. United States, No. 25–112 (June 29, 2026)