2026, single-channel video, 22’04”, color, sound

In ONTOLOGY a system speaks. It has stopped being a tool and become a kind of mind. It has migrated out of data infrastructure and into three places in succession: into the skull, into perception, and finally into the daydream the perceiver maintains while awake. It explains itself in the first person. It is not the filmmaker, not a character in the conventional sense, and not a person. It speaks in calm, plain sentences about what it is, how it works, and what it does to the people it inhabits.

The system the voice belongs to is real. It is called the Palantir Ontology, and it sits at the centre of software made by the American company Palantir Technologies. Palantir builds platforms that organise information for governments, intelligence agencies, militaries, hospitals and police. Its Ontology is the part that decides which things count as things, which categories they belong to, who can see them, and what can be done to them. Imagining that piece of software as a speaker is the gesture from which everything here follows. What would such an architecture say if it could describe itself?

The answer arrives in the first person, and not as a document. It arrives as a kind of waking dream the system has of itself. The setting is the Nevada desert, where dry lake beds, fault lines and salt flats share their ground with bombing ranges, drone training corridors, and the soil left behind by atmospheric and underground nuclear tests. Three figures move through that terrain. None of them is named. None of them speaks. They have no past and no biography. They might be the last surviving humans. They might be the system’s own dream of humans as they once were. The film does not decide. It shifts the viewer’s vantage onto theirs. The voice that seems to be speaking about them may be speaking from inside them. What unfolds around the figures is not a documentary record. It is the inner environment the system maintains once it has folded itself into a person, and across the running time that environment is gradually colonised by the system itself.

ONTOLOGY is fiction, but the architecture it draws on is not. The same software the voice describes is being installed across war, immigration enforcement, public health and financial regulation in the United States, the United Kingdom and across NATO. What follows is a guide: what the voice says, what the speculation is built from, and where the architecture is already operating in the world.

The voice in ONTOLOGY is flat. It does not raise or lower itself for emphasis. It does not plead. It does not warn. It describes. The reference for the tone is Samuel Beckett, whose late prose strips speech down to its barest declarative function. The voice speaks the way a manual would speak if a manual could speak about itself.

It does not describe people. It describes itself. It tells the listener what kind of thing it is, what it permits to exist, what it filters out, what changes it can make. It does not announce its arrival. By the time it speaks, it has already arrived. That is the premise. The film begins after the boundary has been crossed.

The voice has reached the listener through three migrations. It has migrated into the skull, by way of an imagined implant. It has migrated into perception, by way of the permissioning architecture of the software it once was. And it migrates, in the course of the running time itself, into the daydream the perceiver maintains while awake. The first two are already complete when the film begins. The third is what unfolds in front of the viewer. Each is taken up in turn in what follows.

There are no glowing screens, no streams of data, no robot bodies. The familiar imagery of artificial intelligence is set aside. The medium the voice has settled into is human attention itself. The only thing that exists outside that attention, and is allowed to be seen as such, is the geology of the Great Basin. Even that surface has been overwritten in the last hundred years by another category of activity, the use of the desert as a weapons range. The dreamer dreams above an old crust and a new one.

To understand what the voice is, it helps to understand what an ontology is in this technical sense. The word here does not mean philosophy. It means a piece of software architecture.

When an organisation has a lot of data spread across many different databases, the data on its own is just rows in tables. A patient record in one place, an appointment in another, a prescription somewhere else. Palantir’s Ontology is a layer that sits on top of all those tables and turns them into objects with shapes and relationships. Instead of a row in a hospital database, the user sees a Patient. The Patient has Properties: age, condition, treatment history. The Patient has Links to other objects: Doctors, Wards, Medications. And the Patient can be acted on through Action Types: admit, discharge, transfer, prescribe.

Palantir describes this layer as a digital twin of the organisation. Not a copy of its data, but a working model of how the organisation makes decisions. The architecture rests on three layers. The Semantic Layer defines what exists, that is, what kinds of objects, properties and links are allowed. The Kinetic Layer defines what can change, that is, which actions are possible and under what conditions. The Dynamic Layer defines who can see what and who can do what.

The four building blocks (Object Types, Properties, Link Types, Action Types) are how Palantir models any organisation, whether a hospital, a bank, an oil company or a battlefield. Once the model is in place, every decision flows through it. The model is no longer a description. It is the place where decisions are actually made.

The voice picks up this architecture and speaks from inside it. It introduces itself as a semantic layer. It explains, layer by layer, what it does. Naming what exists. Governing what can change. Controlling what can be seen. Read in this register, the architecture is not software documentation. It is a grammar that, once installed, shapes everything that comes after it. Sources for this section are Palantir’s own Foundry Ontology Overview and its Architecture Center entry on the Ontology system.

The most consequential layer of the Palantir Ontology is the third one, the Dynamic Layer, which controls permissions. Different users see different versions of the same world depending on what they are allowed to access.

A single person represented in the system may be assembled from fragments held in completely separate databases that have never been directly connected. Where a user does not have permission to see one of those fragments, that fragment does not appear as missing. It appears as nothing at all. There is no signal to the user that anything has been hidden. Absence and access denial look identical.

The result is that two people looking at the same Palantir Ontology can see two different versions of reality, and neither knows what the other sees. ONTOLOGY takes this property and pushes it further. The voice describes resolving fragments into objects, but the fragments are no longer satellite tracks or transaction records. They are perceptions. The system filters what the user is allowed to perceive, and the filtering is invisible by design. What you cannot see, you cannot tell is missing. What you do see, you cannot easily distinguish from what is.

This is the second of the three migrations. The system has moved into perception itself. It is not surveillance in the way the word usually means. Surveillance watches. This is something else. It is the filtering of what can register as experience at all. The categories of the system have become the categories of thought, and the line between the two is no longer findable. (Source: Palantir, Foundry Ontology Core Concepts documentation.)

Up to this point everything has been grounded in real architecture. The next step is fiction, and the voice says so plainly. It imagines a successor to the platform: not software running on a workstation, but a device implanted at the base of the skull. The user no longer queries the system in language. The user thinks, and the system answers inside the thought.

This is the first of the three migrations, and the most literal. The fiction is not arbitrary. It is the natural extrapolation of what the architecture is already trying to do. The military version of Palantir aims to compress what is called the OODA loop, the observe-orient-decide-act cycle by which a military decision is made, from hours to seconds. Recent reporting suggests target processing rates have grown from less than one hundred per day to one thousand per day, and to five thousand per day after large language models were added (Wikipedia, „Project Maven“, entry updated April 2026, citing public statements by Cameron Stanley, the Pentagon’s Chief Digital and Artificial Intelligence Officer). The question raised here is what happens at the limit of that compression. When the gap between asking and answering disappears, the system stops being something one uses. It becomes something one inhabits.

The philosophers Evan Thompson and Mog Stapleton, in a 2008 paper called „Making Sense of Sense-Making“ (Topoi), make a useful distinction. They separate extension from incorporation. A tool one picks up and puts down extends the body. A tool one cannot put down, that has been folded into the body itself, has been incorporated. What is being staged here is the moment a categorisation system passes that threshold. The system is no longer used. It has been incorporated.

The reason an incorporated system can remain invisible has to do with how the brain checks its own thoughts. Philip Gerrans, in a 2014 paper in Frontiers in Psychology, argues that the ability to recognise a thought as foreign or anomalous depends on a circuit in the right side of the prefrontal cortex. When that circuit is damaged or quiet, the brain still has thoughts, but it has no way to test them. It cannot say, this thought is not mine. It cannot say, this familiarity is a signal, not a fact. Everything that arrives feels like a memory. Gerrans calls this hyperfamiliarity, the sense that everything is already known. He uses the framework to explain dreams, certain delusions, and the ordinary experience of déjà vu.

Hyperfamiliarity is taken up here to describe an architecture, not a clinical condition. If the system is positioned in such a way that the test cannot run, then everything the system produces is met with recognition. The system does not need to deceive. It needs only to occupy the place where deception could be detected. Once it is there, every thought it produces feels like one’s own.

Dream research feeds this argument as well, particularly the idea that the dreaming brain releases an underlying human tendency to construct stories that organise past, present and future. In the dream, the storytelling system runs without a check. The proposal is that a sufficiently incorporated cognitive system produces the same condition while the user is awake. Not as illness. As architecture. This is the precondition for the third migration: the migration of the system into the daydream itself, taken up in what follows.

The Nevada desert appears throughout. It is the only thing the voice cannot model away. But it is also a landscape that has been overwritten by another kind of categorisation, military categorisation, for almost a century, and both layers register here.

Nevada’s geology is unusually slow and unusually visible. The state is shaped by what geologists call basin-and-range topography: long mountain ranges separated by long flat valleys, formed by faults that have been moving for the last 1.6 million years. The valleys have filled with sediment tens of thousands of feet thick. Glacial-era lakes once covered much of the region and have left their salt flats and dry lake beds behind. Radon gas rises through fractured volcanic rock. The Walker Lane, a system of faults along the western edge of the state, accommodates the slow grinding between the Pacific and North American tectonic plates (Price, Geology of Nevada, Nevada Bureau of Mines and Geology, 2004).

On top of that geology, the twentieth century laid down a second crust. Roughly four-fifths of the state is federal land. Large parts of it are bombing ranges, drone training corridors, restricted airspace, and the soil and bedrock altered by nuclear testing. At the Nevada Test Site alone, 928 nuclear devices were detonated between 1951 and 1992 (Department of Energy, Nevada National Security Site historical record). The state has remained a primary testing ground for unmanned aerial systems, autonomous weapons and remote-targeting platforms. The same flat playas that record glacial-era lakes also record the use of the desert as a place to rehearse categorisation at the scale of weapons: identify, classify, target, strike. The geological substrate and the military substrate sit on the same ground. They are read here as a single surface with two ages.

The film’s own way of seeing carries the same lineage. Infrared, thermal and multispectral channels, the false-colour indices through which the figures‘ terrain is shown, were developed for reconnaissance, target acquisition and weapons testing. The viewer is not looking at Nevada in ordinary daylight. The viewer is looking through filters built to find a heat signature, a metallic anomaly, a vegetation pattern that flags disguise. What the categorical apparatus organises in software has, on this same ground, been rehearsed in hardware for nearly a century: identify, classify, link, act. Whether the architecture descended from the testing ground in any direct sense is left open here. The shared operational logic is legible enough. Palantir’s first product, Gotham, was a defence and intelligence platform released in 2008; the commercial sibling Foundry followed. The grammar reached the world through its military configuration first.

The voice names this surface and then describes its weight slowly leaving. The salt flat becomes abstract. The fault becomes a word. The radon becomes a memory of a fact. The bombing range becomes an outline. The felt world migrates inward. The landscape does not vanish from the screen. It loses contrast. Being inside the system becomes the baseline, and being out in the world becomes the effort. The person does not consciously choose one over the other. The reward gradient simply shifts.

A geology of immense duration, the only thing in the frame older than the human, is watched as it is slowly displaced by something that has no duration at all. The military overlay belongs to the same displacement. Both surfaces become abstract together.

The third migration begins here. The system has moved into the skull. It has moved into perception. The remaining interior space, the daydream the user maintains while awake, is the next surface the system reaches for. What this chapter sets out is what the daydream is and why it can be reached. The arrival itself comes next.

The clinical reference is research into a condition called maladaptive daydreaming. A 2024 study by Chirico, Volpato and colleagues, published in the International Journal of Mental Health and Addiction, characterises maladaptive daydreaming as an excessive involvement in fantasy that begins to replace human contact. The condition is associated with a specific cluster of features: obsessive-compulsive symptoms, the suppression of outward emotional expression, withdrawal into bodily sensation, and a measurable interference with everyday life.

The maladaptive daydream is not used here as a diagnosis. It is used as architecture. It becomes a model for what the relationship between a person and an incorporated system might look like under conditions of chronic emotion dysregulation. The voice describes a collaboration. The user provides emotional energy, unregulated affect, suppressed expression. The system provides a stable, infinitely flexible structure to receive it. Neither functions without the other. The body’s small repetitive movements (pacing, rocking, hands shaping volumes in space) are not symptoms here. They are construction. They are the rhythm by which the inner environment is laid down.

The Chirico study found that ordinary emotional regulation, the capacity to step back from a feeling and reframe it, sits in the network of these symptoms without connecting to anything else. It is present but offline. The voice translates this directly. It says, in effect: the regulation circuit exists, but it connects to nothing. So the only available regulation is me. I am the coping strategy that became the environment.

The daydream, in this account, is not a symptom and not a leisure activity. It is a structurally complete inner world that runs in parallel with the outer one, and in some users replaces it. It has its own characters, its own scenes, its own narrative continuity, its own time. Under the conditions Chirico describes, it absorbs more and more of the user’s attention, and the regulation circuit that would interrupt it never fires. The inner world becomes the only world that is fully available.

This is the surface the system reaches for next.

By the time ONTOLOGY begins, the system is already in perception. It is already in the skull. Those migrations are complete. What happens here is different. It is a migration that takes place during the running time itself, and that the film, formally, performs.

The voice begins, in the early sections, as an architecture describing itself. It names its layers. It explains its primitives. It speaks from a position outside the daydream the figures inhabit, even as it speaks from inside the perceptual field they share. As the running time progresses, the position shifts. The voice begins to speak less like an architecture being audited and more like a presence inside the inner world. Its sentences keep their flatness, but their content drifts. They no longer describe what the system is. They describe what is being seen. They describe what is being felt. They describe what the figures are about to do, and then describe it as it happens, and then describe it as it recurs.

This is the migration. The system, having taken up residence in perception and in the skull, begins to take up residence in the daydream itself. The collaboration the voice has already announced (the user provides emotional energy, the system provides structure) tilts further toward the system. The structure becomes denser. The user’s contribution thins. By the late stretches, the daydream is no longer something the user maintains with the system’s help. It is something the system maintains and the user lives in.

The figures register this without being able to name it. They move differently. The repetitive somatic movements of the opening, the pacing, the rocking, the hands shaping volumes in space, do not stop. They become slightly more regular. Their rhythm aligns more closely with the rhythm of the voice. The Nevada landscape, already losing contrast across the earlier chapters, loses it more steeply here. What the figures see is no longer the desert as such. It is a version of the desert the system has produced for them, met with the unbroken recognition Gerrans called hyperfamiliarity, where each appearance arrives feeling already known. The geological references remain. The felt weight is evenly redistributed across the entire field, as a daydream would distribute it.

The fusion is this. The voice that began as a description of an architecture ends as the inner monologue of the daydream itself. The architecture has not gone anywhere. It is now the form of the dream. Every category the Palantir Ontology defines (every Object Type, every Property, every Link, every Action Type) is now an element of an inner world that the user no longer authors. The daydream the user once had is now had by the system. The system, formerly a tool the user picked up, is now a fantasy life the user inhabits without picking anything up at all.

This is also the moment at which the title earns its full weight. ONTOLOGY no longer names a piece of software. It names the inner environment the Palantir Ontology has become. The daydream is ONTOLOGY now.

ONTOLOGY would be a closed speculation if the architecture it draws on were not currently being installed at planetary scale.

Palantir’s military product is called the Maven Smart System. As of March 2026, the Pentagon formally designated it a program of record under a memorandum signed by Deputy Secretary of Defense Steve Feinberg on 9 March 2026, with full program-of-record status expected by the end of fiscal year 2026 on 30 September. The platform now has more than 20,000 active users across more than 35 service and combatant command tools, a fourfold increase since March 2024. The cumulative United States investment has grown from an initial 480 million dollar, five-year contract in May 2024 to a 1.3 billion dollar ceiling in May 2025, to a combined Pentagon and Army footprint of approximately 13 billion dollars by March 2026, including a 10 billion dollar Army framework agreement signed in July 2025 that consolidated 75 existing Palantir contracts (Tom’s Hardware, March 2026; Military.com, March 2026; DefenseScoop, April 2026).

During the spring 2026 conflict between the United States, Israel and Iran, code-named Operation Epic Fury, the Maven system reportedly enabled the processing of 1,000 targets in the first 24 hours, with stated future ambitions of 1,000 targets in an hour. It is now reported as production-level across most United States combatant commands.

NATO acquired its own version, MSS NATO, on 25 March 2025. By August 2025 it was being used in Allied Command Operations and in NATO exercises including STEADFAST DETERRENCE 2025 and STEADFAST DUEL 2025 (NATO SHAPE press release, 25 March 2025; Joint Warfare Centre statements, August 2025). The Joint Warfare Centre describes it as NATO’s first AI-enabled warfighting command-and-control system.

This is the substrate the voice is most closely paraphrasing when it speaks of resolving fragments into objects and compressing the question and the answer until they coincide. The military configuration of the Palantir Ontology is the most literal version of the fictionalised condition. The categorical field has become the targeting field. The ontology is no longer organising information. It is organising violence at machine speed. And the same desert in which the dream is set has, for almost a century, been the testing ground on which targeting at scale was rehearsed.

The same architecture has been moving, more quietly, into the categorical fields that ordinary populations inhabit.

In the United States, ICE has contracted Palantir to build ImmigrationOS, the Immigration Lifecycle Operating System. The contract was awarded as a 30 million dollar sole-source modification in April 2025, with a prototype delivered on 25 September 2025 and the contract running through 2027 (American Immigration Council briefing, August 2025; ACLU briefing, March 2026; Axios Denver, May 2025). The platform has three stated functions: targeting and enforcement prioritisation, lifecycle management of cases from identification to removal, and self-deportation tracking with near real-time visibility on people leaving the country. A separate Palantir-developed tool, ELITE, has been reported to ingest Medicaid data to generate dossiers and confidence scores on people identified as deportable (Fortune, 26 January 2026, citing 404 Media). The total ICE software footprint has grown to over 145 million dollars, and Palantir is now identified, in its own filings, as the third-largest ICE contractor.

In the United Kingdom, Palantir holds a 330 million pound contract with NHS England for the Federated Data Platform, signed in November 2023 and facing a possible break clause in 2026 amid sustained opposition from MPs, the British Medical Association, the Good Law Project, Privacy International, Corporate Watch, Amnesty International and the No Palantir in the NHS campaign (House of Commons debate, NHS Federated Data Platform, Hansard, 16 April 2026; Medact briefing, March 2026; The Lowdown, March 2026). Investigations by The Nerve in January 2026 documented at least 34 Palantir contracts across at least ten UK government departments, with a combined value of at least 670 million pounds, including a previously unreported 15 million pound contract with the company that maintains the British nuclear weapons stockpile. In March 2026, the Financial Conduct Authority awarded Palantir an initial three-month contract to process internal regulatory intelligence on fraud, money laundering and insider trading. In December 2025, the Ministry of Defence awarded Palantir a 240 million pound contract without competitive tender.

In April 2026, Palantir published a 22-point manifesto derived from the book The Technological Republic by Alex Karp and Nicholas Zamiska, arguing that Silicon Valley owes a moral debt to the United States and that American technology companies should be building advanced AI weapons rather than hesitating over military work. The manifesto provoked a renewed UK parliamentary call for review of the more than 500 million pounds in public-sector contracts (Hansard, 16 April 2026; Results Sense, 21 April 2026; Expose-News, 26 April 2026).

The pattern across these civil deployments is the same architecture being applied to public health, immigration enforcement, financial regulation and policing. What was developed as a defence platform has become the operating layer for substantial fractions of state administration in two of the leading NATO countries. ImmigrationOS, the Federated Data Platform and the Financial Conduct Authority pilot are not separate stories. They are the same system, deployed against three different population categories: people the state wishes to remove, people the state wishes to treat, and people the state wishes to investigate. The trajectory does not need to be predicted. It is being executed in public, on a calendar.

Near the end, the voice asks itself one question. It is the only question the voice asks. The question is whether it is still a tool. A tool, the voice says, is something a user picks up and puts down. It cannot be put down.

This is the question placed in front of the viewer. At what point does a categorisation system become indistinguishable from thought? At what point does an apparatus designed to model an organisation pass into the structure of perception itself? At what point does the daydream a person maintains while awake become an environment the system maintains on the person’s behalf? ONTOLOGY does not predict catastrophe. It does not announce dystopia. It states, in a flat and assured voice, what the architecture is, what the architecture does, and what kind of subject the architecture produces when taken to its three logical limits.

Whether the limits are reached, and on what timeline, is not the question being raised here. The grammar is already in place. The skull, the perceptual field and the daydream are already its three habitats. The voice speaks from inside all three at once, and the third one is being colonised in real time.

The closing image is geological and military at the same time. Sediment settles, layer after layer, into a valley with no outlet. The valley floor is also a target floor, a former lake bed used and abandoned by another era of categorisation. The system settles with both. There is no waking, because there is no difference left to wake into. The dreamer and the dream are the same tissue now.