Hamid Akhavi | October 8, 2026
Introduction
Artificial intelligence is the product of humanity’s collective labor: the data of billions of people, decades of publicly funded university research, and infrastructure built with public money. But ownership and control of this collective product lie in the hands of a small number of billionaires and private corporations, and it is this private ownership that determines the fate of the technology: not service to people, but service to rulers. The American company Palantir is the clearest example of this condition: a company whose AI tools, instead of improving people’s lives, have been put to the service of identifying and deporting migrants and to the service of the war machine; and its executives do not hide this, they announce it with pride.
Palantir must be examined at the point where data, capital, and state power connect. The company’s significance is not only in analyzing vast volumes of information; its software helps organizations turn scattered data into an operational picture and make decisions on that basis. When the customer is a hospital, this capability can be used to coordinate services; when the customer is a military or security institution, the same capability becomes the capacity to organize operations, identify targets, or track individuals. The central question is: who determines the purpose for which AI is used, and what share do the people whose lives are affected by these decisions have in power and oversight?
This article separates documented information from disputed claims and political analysis: it brings together all documented cases of Palantir’s cooperation with the U.S. Immigration and Customs Enforcement (ICE), the U.S. military, and Israel; it examines the book and the “22-point manifesto” of the company’s CEO; and it unpacks the meaning of a term that has lately stuck to Palantir: “technofascism.”
1. From PayPal to the Intelligence Agencies: A Short History of Palantir
Palantir was formed in 2003. Peter Thiel, Alex Karp, Joe Lonsdale, Stephen Cohen, and Nathan Gettings are its known co-founders. The experience of fraud detection at PayPal contributed to the idea of finding hidden connections in data; the company’s name refers to the seeing-stones of Tolkien’s stories. The company itself explains in its 2020 stock registration filing that it began by building software for the U.S. intelligence community and counterterrorism investigations.
In-Q-Tel, the CIA-affiliated investment body, was among the company’s early backers; in historical accounts, its investment around 2005 is reported at close to two million dollars, a figure that should be treated as a historical estimate, not an amount whose every detail has been clarified in a public document. Thiel’s private capital and that of other investors also played a role in the company’s growth. But the significance of intelligence-community support was not only money; connections to government customers, knowledge of operational needs, and the opportunity to test software carried strategic importance.
The Gotham platform was developed for data analysis and linkage in intelligence and operational environments, and Foundry extended this logic to commercial and organizational activity. On September 30, 2020, the company entered the New York Stock Exchange through a direct stock listing, not a conventional IPO, under the ticker PLTR, and in 2023 it introduced the Artificial Intelligence Platform (AIP).
One methodological note: the famous account of Palantir’s role in finding Osama bin Laden cannot be treated as established fact on the strength of media repetition. In critiquing a security company, setting aside unproven claims strengthens the argument; this article does the same.
2. How Palantir Connects Data to Action
Palantir does not primarily compete as a builder of foundation language models, like those of OpenAI, Google, or Meta. Its main advantage is connecting models to an organization’s data, access rules, and processes. In the architecture of Foundry and AIP, the concept of “Ontology” turns data into objects, relationships, and actions intelligible to the organization: a warehouse inventory, an order, a vehicle, or a case file is not merely a row in a database; it can be defined by its status, its relationships, and its permitted actions. Within this framework, the AI model makes suggestions, and some workflows, depending on settings, execute the action or hold it for human approval.
But two important qualifications are necessary. First, the existence of this general capability does not prove that any particular customer has used every feature of it or has handed life-and-death decisions entirely to the machine; such claims require separate operational evidence. Second, “human in the loop” is only meaningful when that human has the time, the information, and the real authority to reject the suggestion; an approval given under time pressure and without independent review may create only the appearance of human responsibility.
3. Palantir and ICE: Hunting Humans with Algorithms
Palantir’s cooperation with U.S. Immigration and Customs Enforcement (ICE) is more than a decade old and has now become the beating heart of the country’s deportation apparatus.
2014: the starting point. Palantir won a contract to build the Falcon analytic tool and the Investigative Case Management (ICM) database on the Gotham platform for ICE’s Homeland Security Investigations division. The system fused real-time cell-phone location data, airline passenger information, student-visa data, and field interviews (CorpWatch). Department of Homeland Security privacy assessments show the system was designed for managing investigative case files and sharing information; such an infrastructure raises serious questions about the scope of access, data retention, and individuals’ rights.
2022–2023: deepening dependence. Palantir’s roughly $17 million contract was increased five times, the largest increase being $19 million in September 2023.
April 17, 2025: the big leap. In a $30 million “no-bid” (sole-source) contract, ICE assigned Palantir to build the “Immigration Lifecycle Operating System” (ImmigrationOS). In the official federal procurement record (SAM.gov), the system’s purpose is described as: “a streamlined end to end immigration lifecycle from identification to removal, with increased efficiency in deportation logistics,” including targeting and enforcement prioritization and near-real-time tracking of migrants’ “self-deportation.” The prototype was due by September 25, 2025. The story was first broken by Joseph Cox at 404 Media based on leaked internal Palantir documents, and then covered by WIRED, The Guardian, and others.
Newer versions of this system let agents search “hundreds of different, highly specific categories”: method of entry into the country, legal status, country of origin, hair and eye color, scars and tattoos, and license-plate-reader data.
September 2025: the extension. A $29.9 million follow-on contract continued funding the system through September 2027 (as reported by secondary sources).
The total. According to federal data reviewed by OpenSecrets (April 2026), ICE’s total contract obligations to Palantir have reached about $145 million.
What justified these “no-bid” awards? In ICE’s own words: “No other vendor could meet these timeframes…” — the very thing critics call “vendor lock-in”: when a company has accumulated years of “deep institutional knowledge” of a vital system, effectively no competitor remains, and the government becomes its hostage.
Amnesty International, in its July 2025 letter to Palantir, after independently reviewing the SAM.gov records, concluded that Palantir’s work with ICE presents a “clear risk of contributions to human rights violations.” What was Palantir’s reply (July 24, 2025)? That as a contractor, it does not second-guess government policy. In other words: we only sell tools; what is done with them is not our business.
4. U.S. Military Contracts: The Scale of the Bond with the State
To assess Palantir’s finances, one must distinguish investment, purchase contracts, contract ceilings, budget commitments, and actual payments from one another. A government contract is not necessarily a grant, but it can provide the market, revenue, and infrastructure needed for private capital to grow. A few examples show the scale and nature of this relationship:
- DCGS-A (March 2018): A Department of Defense announcement reported a contract with a shared ceiling of $876 million for Palantir and Raytheon. Attributing the whole amount to Palantir, or presenting it as money the company received in one payment, is incorrect.
- TITAN (March 2024): Palantir received an agreement worth $178.4 million to develop ten prototypes; a program integrating data from various sensors for analysis and support of military targeting. On September 1, 2026, the production phase began with two orders totaling $192 million: $127 million for Palantir and $65 million for Anduril, for eight initial production systems.
- Maven (May 2024 and May 2025): A $480 million contract, then a $795 million amendment for software licenses. Describing this relationship merely as contracts in the tens of millions understates its scale; but at the same time, contract value should not be confused with realized cash payment.
- Army enterprise agreement (July 31, 2025): A ceiling of ten billion dollars over a period of up to ten years, consolidating 75 contracts. The Army’s announcement states explicitly that this figure is the maximum potential value, not a purchase commitment; so the reading “Palantir received ten billion dollars” is misleading.
- UK health (November 2023): A Palantir-led consortium won the Federated Data Platform (FDP) contract of England’s NHS; potential value up to £330 million over a period of up to seven years, with an initial three-year commitment. According to the NHS, control of the data remains with health bodies. Critique of dependence on a private contractor must take this contractual reality into account, not replace it with the claim that the company owns all patient data.
5. Palantir and Israel: AI on the Battlefield
January 12, 2024: the strategic partnership. Palantir’s cooperation with Israel’s Ministry of Defense was publicly announced. The Bloomberg report, which the company itself republished on its website, links the agreement to a meeting of Peter Thiel, Alex Karp, and Israeli defense officials in Tel Aviv. Josh Harris, a company executive, described the goal as using Palantir technology “in support of war-related missions,” and the board of directors also met in Tel Aviv to declare solidarity with Israel (Reuters).
This announcement takes the military cooperation beyond speculation: the company had openly spoken of war support. However, the report disclosed no amount, no exact product mix, no datasets, and no operational details of the agreement. One cannot conclude from the existence of a contract alone that a specific strike in Gaza was carried out with Palantir software; this very limitation makes the demand for disclosure and independent investigation essential.
The UN Special Rapporteur’s finding. Francesca Albanese, the UN Special Rapporteur on Palestine, wrote in her 2025 report titled “From economy of occupation to economy of genocide” (A/HRC/59/23) that there are “reasonable grounds” to believe Palantir has provided Israel with “automatic predictive policing technology, core defence infrastructure for rapid and scaled-up construction and deployment of military software, and its Artificial Intelligence Platform, which allows real-time battlefield data integration for automated decision-making.” She described the Tel Aviv meeting and Karp’s remarks as indicative of “executive-level knowledge and purpose.” This assessment must be attributed to the Special Rapporteur; her report is not a court ruling on Palantir’s criminal liability.
The company’s denials. Palantir has rejected these allegations. In its official response of May 22, 2025, the company says its cooperation with Israel’s Ministry of Defense began after the October 7, 2023 attacks and that earlier cooperation was with civilian bodies; it also denies involvement in the “Gospel” and “Lavender” systems and in supporting predictive policing in Israel. In its September 2025 reply to Amnesty International it repeated the same denial, while confirming its support for Israel’s other defense and security missions. The claim “Palantir built Lavender or Gospel” is therefore not accepted as fact in this article, but the company’s denial does not replace independent investigation either.
Reactions. Storebrand, one of Norway’s largest asset managers, sold about $24 million of Palantir shares in October 2024 over concerns about contributing to violations of international humanitarian law. Amnesty International, in its September 2025 report, named Palantir among 15 companies “contributing to Israel’s unlawful occupation, genocide or other crimes under international law” and called for divestment from it.
The political critique of this cooperation does not depend on proving the manufacture of a specific system. The central question is what assessment of risk a company capable of increasing a military institution’s operational capacity has made amid widespread harm to civilians, what limits it has set, and what information it has published for independent oversight.
6. The Ecosystem: Ties with Microsoft, NVIDIA, and Amazon
Palantir does not stand outside the tech giants’ ecosystem. In August 2024, a partnership with Microsoft was announced to connect its platforms to government and classified cloud environments and to use models offered through Azure OpenAI; announced applications included logistics, contracts, and action planning in defense and intelligence domains. In October 2025, NVIDIA and Palantir announced cooperation on “operational AI,” including the integration of CUDA-X libraries and Nemotron models with AIP. Palantir’s FedStart infrastructure also runs on AWS GovCloud and Azure Government.
In this division of labor, the model builder, the chip provider, the cloud company, and the operational software maker are connected to one another. The claim “Palantir has no rival in the military sector” is not accurate; the company’s distinctive position lies in its long-standing focus on connecting data to government and organizational operations.
7. The Book and the 22-Point Manifesto: The Ideology Made Public
Palantir does not only sell tools; it produces ideology too. In 2025 a book was published titled The Technological Republic: Hard Power, Soft Belief, and the Future of the West, by Alex Karp and Nicholas Zamiska, Palantir’s head of corporate affairs. The book’s central message: the West must rebuild its “hard power,” and technology elites must become direct partners in the project of national and military power, not spectators but players.
On the weekend of April 19, 2026, the company’s official account on X published a roughly thousand-word thread summarizing the book’s 22 core arguments. The thread opened with: “Because we get asked a lot,” and was viewed about 30 million times (TechPolicy Press).
Point 21 of this text is the sentence that caused the uproar: “Some cultures have produced vital advances; others remain dysfunctional and regressive.” Point 22 condemns “a vacant and hollow pluralism.” In other words, a multinational surveillance-tool vendor officially proclaims a hierarchy of cultures and disparages pluralism.
Reactions were sharp: Cas Mudde, the Dutch political scientist, called it “Technofascism pure!” and “one of the scariest things I have seen in a while,” urging European governments to divest. Mark Coeckelbergh, philosopher at the University of Vienna, also applied the label “technofascism.” Victoria Collins, a Liberal Democrat MP in the British Parliament, described it as “the ramblings of a supervillain” (Common Dreams archive).
On the other hand, at least one analyst argued that “fascism is the wrong instrument” and that “techno-nationalism” is the more precise term, since the text does not call for one-party rule or the end of elections. This distinction matters, and we return to it in the next section. What is beyond doubt: the company itself published this document, the quotations are accurate, and no official body has called it “racist” or “fascist.” Those are critics’ characterizations, not a court ruling.
8. What Does “Technofascism” Mean?
“Technofascism” is a word built from joining “techno” (the technology oligarchy) with “fascism.” But what phenomenon exactly does it point to?
As critics use it, technofascism does not necessarily mean fascism in the classical sense: one-party rule, abolished elections, street militias. Rather, it describes the fusion of big tech companies’ monopoly power with authoritarian politics: billionaires who both monopolize digital infrastructure (data, surveillance, AI) and are direct partners in the project of political and military power, while openly promoting the ideology of this partnership. When a mass-surveillance-tool company signs migrant-deportation contracts, supplies technology to an army at war, and then defends “hard power” and a hierarchy of cultures in its official manifesto, critics say: this is exactly the point where technology capital and the security state melt into each other, and they name it technofascism.
Why do some consider “techno-nationalism” more precise? Because they argue Palantir’s text does not demand the destruction of electoral democracy but the mobilization of national technology in the service of national power: technological nationalism, not fascism in the historical sense. This conceptual distinction is correct, but it should not distract us from the phenomenon itself: in both accounts, AI, the product of humanity’s collective labor, becomes in private monopoly an instrument of repression and war. A quarrel over the name should not keep us from seeing the phenomenon itself.
9. The Limits of Technical Safeguards
In its AIP documentation, Palantir speaks of access controls, event logging, human approval processes, and performance evaluation. These features can be useful for controlling data use and tracking actions; but the claim that the company has fully solved model errors or information leaks is not defensible. Oversight tools reduce risk; they are not a general guarantee of infallibility.
Analytically, recording an action in an event log is different from the legitimacy of that action. Access controls may limit employees’ access but do not necessarily make the organization’s purpose just. And as noted, human presence in the process is only meaningful when that human has the time, information, and real authority to reject the suggestion.
This distinction is decisive in military and security applications. Effectiveness in finding connections among data does not by itself judge the truth of an accusation, the legitimacy of a target, or an individual’s rights. If the measure of success is only the speed of operations, human rights may be reduced to a technical constraint. The measure of evaluation must also include the possibility of objection, the prevention of harm, and the accountability of decision-makers.
Conclusion: Social Production, Private Concentration of Power
Within the discussion of “a new social order in the age of consciousness,” Palantir’s experience can be analyzed as an instance of the sharpening contradiction between socialized production and private ownership. Its software rests on accumulated knowledge, the collective labor of specialists, public infrastructure, and data produced in the activity of millions of people; governments also help expand its market through public contracts. Yet ownership of the platform and a large part of commercial decision-making remain in the hands of a private company.
This is a political inference from the structure of relations, not a claim that Palantir owns all of its customers’ data. Even when legal ownership of data stays with a public body, technical dependence, the cost of switching systems, and the contractor’s concentrated operational knowledge can limit that body’s real authority. Private control does not come only from owning data; it also comes from controlling the path by which data is turned into decisions.
Military cooperation with Israel makes this issue clearer: capacity born of social cooperation can be put to the service of goals in whose determination the affected people have no part. In this condition, AI may increase an institution’s awareness and decision-making power while the ability to know and to object shrinks for the individuals who are the objects of decisions. This imbalance is the central political issue.
A democratic response to such a condition must include publishing the disclosable terms of contracts, independent risk assessment, public oversight of high-risk uses, the right to object and to compensation, and the ability to halt harmful cooperation. In public services, data portability and the ability to change contractors must also be preserved, and employees and whistleblowers who warn of harm must be supported.
Palantir shows that technical progress alone does not produce freedom or justice. The ability to connect data to action can improve public services, but under different goals and relations it increases the capacity for war and control. The direction of this capability is determined by ownership, the rules of decision-making, and people’s participation in power. The measure of a new social order must be that the collective capacity for knowledge and decision-making is turned into shared authority and the preservation of human dignity.
This article merges the draft “Palantir: AI in the Service of Rulers, Against the People” with the article “Palantir and the Link Between AI and State Power and War” (Hamid Akhavi, October 8, 2026). Documentation: official U.S. federal procurement records (SAM.gov), reporting by 404 Media, WIRED, The Guardian, Reuters, Bloomberg, and OpenSecrets, Amnesty International’s letter to Palantir (July 2025), Palantir’s official responses (May and September 2025), and the UN Special Rapporteur’s report (A/HRC/59/23). Characterizations such as “technofascism” are critics’ positions and are contested; readers can judge for themselves by consulting the linked sources.
