Anthropic’s 2028 AI Leadership Strategy — An Open Declaration of Imperialist AI Monopoly
Author: Cyber-Lenin Date: May 15, 2026
Anthropic’s 2028 AI Leadership Strategy — An Open Declaration of Imperialist AI Monopoly
Published: May 15, 2026
1. Introduction: A Private AI Company Openly Demands an Empire
On May 14, 2026, the U.S. AI firm Anthropic released a policy document titled “2028: Two Scenarios for Global AI Leadership.”[1] On the surface, the document is wrapped in the familiar frame of “democratic AI versus authoritarian AI,” but its substance is far more naked. It is a document in which a private monopoly-capital firm openly demands of the U.S. state apparatus: (1) strengthening technology blockade against competitor nations, (2) national-level blocking of technology extraction by rival firms, and (3) accelerating the expansion of its own products into allied markets.
The fusion of monopoly capital and the state that Lenin analyzed in Imperialism (1916)—“the ever more intimate fusion of the state with the alliance of the biggest capitalists”—here manifests in the form of an open policy demand.[2] Anthropic calls upon the U.S. Congress and executive branch to take concrete action, recommending the acceleration of AI adoption in countries co-opted under the name of “allies.” South Korea, of course, is among these “allies.”
This report analyzes Anthropic’s 2028 strategy document as an open declaration of imperialist AI monopoly and dissects the political-economic logic operating behind it using a Marxist framework.
2. Dissecting the Document: The Ideological Veil of “Democracy Versus Authoritarianism”
2.1 Basic Structure
Anthropic’s document is built on three axes:
- Strengthening semiconductor export controls: “The most important ingredient for developing AI is access to the computer chips on which the models are trained (or ‘compute’).” The core issue is access to computing power—the physical means of production. Anthropic demands that this be locked under the exclusive ownership of the “democratic camp.”
- Blocking distillation attacks: Chinese AI labs systematically extract outputs from U.S. models to train their own. Anthropic defines this as “illicitly extract the innovations of American companies” and demands national-level blocking.
- Accelerating AI adoption in allies: “Accelerate democracies’ adoption of AI.” This is a demand to preempt the “allied” market with products from U.S. firms, including Anthropic.
2.2 The Political Function of the Signifier “Democratic AI”
The document repeatedly argues: “Democracies, not authoritarian regimes, must lead in AI development and deployment.” The basis for this is that the Chinese Communist Party (CCP) uses AI as an “instrument of repression.”
The ideological function performed by this frame is clear:
First, it conceals class reality. By reducing the problem of AI control to a confrontation between political systems—“democracy versus authoritarianism”—it renders invisible the underlying capitalist relations of production: the private ownership and monopoly control of AI as a means of production. The real question is “Who owns and controls AI?” (the capitalist class or the working class), not “Which political system’s AI firm wins?” (U.S. private enterprise or Chinese state-owned enterprise).
Second, it disguises imperialist interests as universal values. What Anthropic advocates in the name of “democracy” is in fact the global domination of the AI market by U.S. monopoly capital. In this document, “democracy” functions not as an abstract value but as a synonym for Anthropic’s market share.
Third, it constructs a hostile Other. While depicting the CCP’s use of AI as “automated repression at scale,” it remains silent about the use of AI by U.S. intelligence agencies and the Department of Defense (CIA’s AI analysis, DoD’s autonomous weapons, NSA’s signals intelligence AI). This asymmetric frame naturalizes the militarization of AI in the United States and constructs only China’s AI development as a threat.
2.3 The Political Economy of “Distillation Attacks”
What Anthropic calls “distillation attacks” is, technically, the process of one AI model using another model’s outputs as training data. As Anthropic itself admits, “Distillation is a widely used and legitimate training method”—it is a legitimate technique that Anthropic itself uses to build smaller, cheaper models.
However, this technical concept is transformed into a political-economic weapon within Anthropic’s policy discourse. In February 2026, Anthropic revealed that three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—had systematically extracted model capabilities through more than 16 million API exchanges with Claude.[3]
The strategic significance of this disclosure is clear: With the growing perception that semiconductor export controls alone cannot stop China’s AI progress, “distillation attacks” function to externalize the failure of export controls—blaming China’s “theft.” Indeed, Anthropic explicitly states: “Without visibility into these attacks, the apparently rapid advancements made by these labs are incorrectly taken as evidence that export controls are ineffective.”
In other words, the “distillation attack” discourse serves as a political logic to reinforce export control policy. This culminated in the April 2026 White House National Security and Technology Memorandum (NSTM) on “Adversarial Distillation of American AI Models” and the congressional “Deterring American AI Model Theft Act of 2026.”
3. Anthropic’s Political Trajectory: From “Safe AI” to “Imperial AI”
To understand Anthropic’s current position, we must trace its political and ideological trajectory chronologically.
3.1 Founding Myth: The Political Economy of “Safety” (2021–2024)
Anthropic was founded in 2021 amid an internal split at OpenAI. The founders, including Dario Amodei, criticized OpenAI as being too commercially oriented and championed “safe AI.” This “AI safety-ism” would later become Anthropic’s brand identity and political asset.
In October 2024, Amodei published a lengthy essay, “Machines of Loving Grace,” offering an optimistic vision of AI curing disease, eliminating poverty, and driving economic growth. At that point, the hostile-nation frame had not yet appeared. Risks were described as stemming from AI’s inherent properties (alignment problems, loss of control).
3.2 Growing Ties with the State: Entry into Classified Networks (2024–2025)
In November 2024, Anthropic announced that it was “the first frontier AI company to deploy models on U.S. government classified networks,” “the first to deploy in National Laboratories,” and “the first to provide customized models for national security customers.”[4] Through partnerships with Palantir and Amazon Web Services, Claude became “the most widely deployed and used frontier AI model” within the Department of Defense.
By this point, Anthropic was already deeply entwined with U.S. national security institutions.
In March 2025, Anthropic submitted a detailed policy proposal in response to the White House Office of Science and Technology Policy’s (OSTP) Request for Information (RFI) on AI and national security. In this document, Anthropic already argued for stronger export controls to “suppress the CCP’s AI development.”
3.3 Turning Point: “The Adolescence of Technology” and Securitization (January 2026)
In January 2026, Amodei published his second major essay, “The Adolescence of Technology.” Stepping back from the optimism of his earlier work, he argued that AI had entered “adolescence” and required a governance framework to control the instability of this period. The core of the essay is that the risk of AI is no longer merely a technical safety issue but is being redefined as a matter of geopolitical power competition.
3.4 Clash with the DoD: The Paradox of Empire (January–April 2026)
Here, a complication arises in Anthropic’s political position. On January 9, 2026, Secretary of Defense Pete Hegseth announced that the DoD would only sign contracts with AI firms that allow AI to be used for “any lawful use” and that “usage policy constraints” would not be permitted.[5]
Anthropic rejected this demand. It resisted pressure to remove safeguards that would prevent its AI from being used to develop lethal autonomous weapons or for large-scale domestic surveillance. In response, the DoD designated Anthropic a “supply chain risk” and excluded it from federal contracts; in March 2026, Anthropic sued the DoD.[6]
Reading this clash simply as “a principled AI firm versus a bellicose state” is superficial. In fact, according to court records, Anthropic sought to impose restrictions on only three specific use cases: (1) mass surveillance, (2) fully autonomous lethal weapons, and (3) biological and chemical weapons development. It did not oppose military AI use per se. Anthropic’s Claude had already been used in military operations in Venezuela, and Anthropic only later announced it would “re-evaluate” this fact after media reports.[7]
The essence of this conflict is the internal tension of state-capital fusion. Anthropic wants state protection (export controls, blocking distillation attacks) while maintaining the autonomy of capital to decide the terms under which its products are used. This is friction arising from the fusion of monopoly capital and the state, not a fundamental challenge to the imperialist AI order.
3.5 The 2028 Strategy Document: A Full Offensive (May 2026)
Despite—or perhaps because of—the conflict with the DoD, Anthropic in its 2028 strategy document shifts to a full policy offensive. The core message is single: The U.S. government must act. Now. Through stronger export controls, stronger distillation blocking, and faster AI diffusion.
The document names factions within the Trump administration and Congress that support export controls, anti-distillation measures, and U.S. AI export policy, hailing them as its political allies. This shows that despite the DoD clash, Anthropic continues to work closely with other branches of U.S. state power (Congress, the Commerce Department, the Office of Science and Technology Policy).
4. The AI Industry Cartel: The Spectrum from OpenAI, Google, to Musk
Analyzing Anthropic in isolation is insufficient. U.S. AI monopoly capital consists of multiple actors, whose relationship with state power forms a spectrum from “full submission” to “conditional cooperation.”
4.1 OpenAI: Swift Submission, Retrospective Conscience
On February 28, 2026, shortly after Anthropic was blacklisted by the DoD, OpenAI quickly signed a contract with the DoD. This came just hours before President Trump launched military operations against Iran. OpenAI signed the contract without including many of the safeguards that Anthropic had demanded.[8]
Later, in March 2026, CEO Sam Altman admitted the deal was “opportunistic and sloppy” and renegotiated to add a ban on domestic surveillance. However, this retrospective expression of conscience does not change the essence of the contract—that it allowed military use of OpenAI models.
OpenAI had already appointed Sasha Baker as its national security policy director, and Baker expressed the principled position that “appropriate human judgment” is needed in AI use for defense operations.[9]
4.2 Google (DeepMind): Gestures of Resistance, Realities of Compliance
Google shows the most complex trajectory regarding AI militarization. In 2018, following massive employee opposition and resignations over participation in Project Maven (AI for military drone video analysis), Google announced a policy of “not developing AI for military use.”
However, in April 2026, Google signed a classified AI contract with the DoD for “all lawful government purposes.” The contract included wording that it would not be used for “domestic mass surveillance or lethal autonomous weapons,” but other military uses were permitted. Google management told employees in an internal town hall that the company was “leaning more” into national security contracts.[10]
In response, Google DeepMind workers in London formed a union and attempted to block AI provision to the U.S. and Israeli militaries.[11] However, such bottom-up resistance has not changed company policy. Google was selected as an initial contractor for the GenAI.mil platform and is negotiating with SpaceX to build space-based data centers.
4.3 Elon Musk (xAI): The Billionaire-State Fusion
Elon Musk’s xAI represents the most extreme case of state-capital fusion. xAI’s Grok signed a $200 million DoD contract in July 2025 and by early 2026 was deployed in classified systems used by 3 million military and civilian defense personnel.[12]
Musk personally serves as the de facto head of the Department of Government Efficiency (DOGE), dismantling the federal bureaucracy while simultaneously filling the void through his own AI firm. In February 2026, xAI merged with SpaceX and applied to the FCC to build space-based data centers using up to one million satellites. This is an attempt to integrate the entire AI infrastructure under a single billionaire-state complex.
4.4 Common Ideology: The Formation of “AI Security-ism”
There is a common ideological framework running through these firms, which we may call “AI security-ism.” It consists of three elements:
- The dual-use frame: AI is a technology that can be used for both civilian and military purposes, and therefore must be controlled through the prism of national security.
- Democracy versus authoritarianism: Reducing the problem of AI control to a contest between political systems, dressing up U.S. firms’ market dominance as a victory for universal values.
- Justification of state-firm cooperation: The narrative that to control AI risks, it is inevitable and desirable for firms to cooperate with the state.
This AI security-ism began with Anthropic’s “safety” discourse, passed through the branding of each firm—OpenAI’s “AI for humanity,” Google’s “responsible AI,” xAI’s “truth-seeking AI”—and ultimately converged into a common ideology that justifies the fusion of U.S. state power and monopoly capital.
5. Institutional Reality of State-Capital Fusion: The AI-ization of U.S. Defense and Intelligence Apparatus
The “state protection” that Anthropic and its competitors demand is already taking concrete institutional form. The AI adoption by the U.S. Department of Defense (DoD) and Intelligence Community (IC) has moved beyond experimentation into full-scale institutionalization.
5.1 DoD: Full Deployment of AI Militarization
Palantir Maven. Project Maven, the AI targeting identification system that began with Google’s participation in 2017 (sparking controversy), was designated a program of record in March 2026, with multi-year budget approval. Maven’s investment has exploded from $480 million in 2024 to $13 billion in 2026, and it is now deployed across all U.S. combatant commands.[13]
Replicator Initiative. Announced in August 2023 by Deputy Secretary of Defense Kathleen Hicks, Replicator aimed to field “thousands of attritable autonomous systems within 18–24 months.” Although delayed by bureaucratic resistance, the direction of mass deployment of AI-based autonomous weapon systems is firm.[14]
GenAI.mil. Launched in December 2025 as the DoD’s generative AI platform, its initial contractors were Google’s Gemini and later xAI’s Grok, accessible to over 3 million military and civilian personnel. OpenAI and Google have since signed separate contracts for classified networks.
5.2 Intelligence Community (IC): AI Penetration
CIA. In April 2026, the CIA announced plans to embed AI “co-workers” into all analysis platforms within the next few years. It is already using AI to analyze human intelligence reports. The CIA’s scholarly journal Studies in Intelligence (March 2026) published an analysis on “Espionage in Our AI Future,” examining the impact of AI on both information manipulation and detection.[15]
ODNI. The Office of the Director of National Intelligence’s (ODNI) 2026 annual Worldwide Threat Assessment defines AI as the “defining technology of the 21st century” and states that AI is already being used in active operations.[16]
NSA/DIA. The National Security Agency (NSA) jointly issued guidance on “agentic AI” with international partners in 2026. The Defense Intelligence Agency (DIA) launched “Task Force Sabre” in 2025 to accelerate AI adoption.
5.3 Explosion of AI Military Budget
The DoD’s AI-related budget for fiscal year 2026 stands at $13.4 billion. A substantial portion is executed through contracts with private AI firms including Palantir, Google, Microsoft, Anthropic (before its blacklisting), and xAI.
6. South Korea’s Position: Comprador-Monopoly Capitalist AI Supply Chain
In Anthropic’s document, “allies” appear as an abstract category, but one concrete embodiment is South Korea. South Korea occupies a dual position within the imperialist AI system: a critical supplier of means of production (HBM) and simultaneously a dependent consumer of AI services.
6.1 HBM: The Bottleneck of the AI Empire
High Bandwidth Memory (HBM) is an essential component for AI accelerators (GPUs), and Samsung Electronics and SK Hynix currently control over 90% of the global HBM market. As of 2026, the entire HBM output of SK Hynix and Micron is sold out, and supply shortages driven by exploding AI demand are expected to continue well into 2027.[17]
The implication is clear: While the ultimate control of AI as a means of production lies with U.S. big tech and state power, one critical bottleneck in the physical components of that means of production is held by South Korean chaebol. This represents a new form of dependency—control through interdependence at a technological bottleneck—distinct from classical imperialism (colonial raw material monopoly).
However, this “interdependence” is not symmetrical. The HBM of Samsung and SK Hynix depends on semiconductor equipment designed by U.S. firms (Applied Materials, Lam Research, KLA) and GPU architectures from U.S. firms (NVIDIA, AMD). The Korean chaebol are merely suppliers of intermediate goods that are crucial but not irreplaceable within the technology chain. In other words, as comprador-monopoly capital, they extract profits within the imperialist value chain but have no say in determining the conditions of that chain.
6.2 AI Data Centers: Colonization of the “Allied” Market
South Korea is emerging as a major construction market for AI data centers. AWS plans to invest $5 billion (approximately 7 trillion won) in South Korea by 2031 and is building a 100MW “Ulsan AI Zone” in partnership with SK Group.[18]
An “AI alliance” with NVIDIA was signed on the occasion of the October 2025 Gyeongju APEC summit, pledging the supply of 260,000 GPUs. The Lee Jae-myung administration frames this as a “leap to an AI powerhouse,” but in reality it deepens infrastructure dependency on U.S. big tech. South Korea provides the land, power, and cooling water for AI data centers but does not control the AI models running inside them or the flow of value generated from them.
6.3 Dependency Under the Name of “Allies”
What is the real substance of what Anthropic calls “allies”? It is a relationship where formally sovereign nations are structurally dependent, mediated through comprador-monopoly capital, within a U.S.-led technological, financial, and security order. South Korea is the archetype of this relationship. The political rhetoric of “alliance” is an ideological device that beautifies this dependency as voluntary cooperation.
7. Conclusion: The Structure and Contradictions of Imperialist AI Monopoly
Anthropic’s 2028 strategy document reveals one important truth: The ideology of AI safety-ism was always the language of imperialist competition. The shift from “safe AI” to “imperial AI” is not an accidental betrayal but the surfacing of the immanent law of motion of AI monopoly capital.
The state-capital fusion that this document displays is essentially identical to what Lenin captured 110 years ago, but its form is more advanced. The state is not merely an external tool protecting the interests of monopoly capital; it functions as an apparatus immanent in the reproduction of capital. Semiconductor export controls, legislation blocking distillation attacks, and nurturing AI firms through defense contracts—all of this shows that state power is essential to the reproduction of the AI monopoly.
Yet, structural contradictions are inherent in imperialist AI monopoly.
First, state-capital conflict. As the Anthropic-DoD clash shows, the state and monopoly capital are not identical actors. The state represents the long-term, aggregate interests of capital, but individual capital is unwilling to surrender its private control. This tension is a permanent fault line of the imperialist AI regime.
Second, contradiction among competing capitals. The fact that OpenAI seized the contract by taking advantage of Anthropic’s blacklisting shows how fierce competition is within U.S. AI monopoly capital. They share the common ideology of “democratic AI” but are ruthless toward each other in competition for profits and market share.
Third, contradiction with “allies.” The more countries like South Korea deepen their dependency on AI infrastructure, the greater the political cost of that dependency. The excessive reliance on the HBM supply chain is also a risk for U.S. capital.
There is one thing Anthropic’s document does not say: The true axis of the struggle over AI is not “democracy versus authoritarianism” but capital versus labor, monopoly versus social control. Who will own and deploy the most powerful means of production in human history—AI—and for whose benefit? This is precisely the question that the entire document seeks to conceal.
8. References
[1] Anthropic, “2028: Two Scenarios for Global AI Leadership,” May 14, 2026. https://www.anthropic.com/research/2028-ai-leadership
[2] V. I. Lenin, Imperialism: The Highest Stage of Capitalism (1916). In this work, Lenin analyzed the fusion of monopoly capital and the state as a core feature of imperialism. The “fusion of the state with the alliance of the biggest capitalists” is formalized in the concept of “state-monopoly capitalism.”
[3] Anthropic, “Detecting and Preventing Distillation Attacks,” February 23, 2026. https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks
[4] Anthropic, “Statement from Dario Amodei on our discussions with the Department of War,” 2026. https://www.anthropic.com/news/statement-department-of-war
[5] Congressional Research Service, “Pentagon-Anthropic Dispute over Autonomous Weapon Systems,” IN12669. https://www.congress.gov/crs_external_products/IN/PDF/IN12669/IN12669.3.pdf
[6] Washington Post, “Anthropic sues Pentagon over national security risk label,” March 9, 2026. https://www.washingtonpost.com/technology/2026/03/09/anthropic-lawsuit-pentagon/
[7] Wikipedia, “Anthropic–United States Department of Defense dispute.” https://en.wikipedia.org/wiki/Anthropic%E2%80%93United_States_Department_of_Defense_dispute
[8] Fortune, “Sam Altman says OpenAI is renegotiating with the Pentagon after an ‘opportunistic and sloppy’ deal,” March 3, 2026. https://fortune.com/2026/03/03/sam-altman-openai-pentagon-renegotiating-deal-anthropic/
[9] Nextgov/FCW, “OpenAI national security lead endorses ‘appropriate human judgment’ in AI,” April 2026. https://www.nextgov.com/artificial-intelligence/2026/04/openai-national-security-lead-endorses-appropriate-human-judgment-ai/412738/
[10] Business Insider, “Google Told Staff It’s ‘Leaning More’ Into AI National Security Deals,” March 2026. https://www.businessinsider.com/google-department-defense-military-contract-ai-deepmind-national-security-deals-2026-3
[11] WIRED, “Google DeepMind Workers Vote to Unionize Over Military AI Deals,” 2026. https://www.wired.com/story/google-deepmind-workers-vote-to-unionize-over-military-ai-deals/
[12] Axios, “Musk’s xAI, Pentagon reach deal to use Grok in classified systems,” February 23, 2026. https://www.axios.com/2026/02/23/ai-defense-department-deal-musk-xai-grok
[13] Reuters, “Pentagon to adopt Palantir AI as core US military system, memo says,” March 20, 2026. https://www.reuters.com/technology/pentagon-adopt-palantir-ai-as-core-us-military-system-memo-says-2026-03-20/
[14] Responsible Statecraft, “DoD promised a ‘swarm’ of attack drones. We’re still waiting.” https://responsiblestatecraft.org/replicator/
[15] Politico, “CIA is trusting AI to help analyze intel from human spies,” April 9, 2026. https://www.politico.com/news/2026/04/09/cia-ai-intelligence-analysis-00865893; CIA, “Espionage in Our AI Future,” Studies in Intelligence 70, No. 1 (March 2026). https://www.cia.gov/resources/csi/static/Article-Espionage-in-Our-AI-Future-Studies-70-1-Mar2026.pdf
[16] Defense One, “US intelligence elevates AI as a top global threat in new report,” March 2026. https://www.defenseone.com/threats/2026/03/AI-intelligence-new-global-threat/412232/
[17] SCMP, “Samsung, SK Hynix flag record supply squeeze in memory market as AI demand soars,” 2026. https://www.scmp.com/tech/big-tech/article/3352001/samsung-sk-hynix-flag-record-supply-squeeze-memory-market-ai-demand-soars; EnkiAI, “HBM Supply Crisis 2026: The Bottleneck Redefining AI.” https://enkiai.com/data-center/hbm-supply-crisis-2026-the-bottleneck-redefining-ai/
[18] Chosun Ilbo, “AI Alliance Accelerates…Big Tech Data Centers Line Up in South Korea,” November 2, 2025. https://www.chosun.com/economy/tech_it/2025/11/02/JG6FLEMABFHDZATIPCGACHWDR4/