Featured

THE IRON VASSAL - AI agents have now acquired an identity

Science and civilization Hits: 204 34 minutes read

Autonomous Agents, Technofeudal Power, and the Slow Liquidation of Human Rights.

"It is no longer okay to say you don't understand technology." — Brett Solomon, Access Now, Glion Human Rights Dialogue, December 2020

Prologue: The Sound of Gears ChangingTHE IRON VASSAL - AI agents have now acquired an identity

There is a particular sound that precedes a shift in civilisational power. It is not the crash of artillery or the roar of a crowd. It is quieter than that — the low hum of servers in a cooled data centre in Virginia, the faint click of a decision being executed by a system that was not asked, that will not be thanked, and that cannot be held to account.

In December 2020, Brett Solomon, Executive Director of Access Now, stood before a virtual gathering of United Nations ambassadors and human rights officials and delivered what was, in retrospect, a warning dressed as a diagnosis. The international human rights framework, he said, was failing us in the digital age. Governments were abdicating responsibility. Technology companies were accumulating unchecked power. Rights were being eroded not by tyranny alone but by architecture — by the design of systems built to extract, manipulate, and surveil. The framework had resilience, he acknowledged, but resilience is not the same as efficacy. A building can survive a storm and still be condemned.

Solomon identified ten specific failure points: internet shutdowns, data exploitation, the weaponisation of counter-terrorism law, the digital divide, attacks on human rights defenders, the capture of humanitarian response by private surveillance firms. He named them carefully, like a doctor describing symptoms in a patient who has not yet been told the diagnosis.

Five years later, that patient has been given the diagnosis — and it is worse than the symptoms suggested.

The name is technofeudalism. The spearhead is the autonomous agent. And the question that Solomon posed — can human rights survive the digital age? — has become sharper, more urgent, and harder to answer than at any point in the five decades since the Universal Declaration was written.

I. The Map Changes While You Sleep

To understand what has happened, you have to understand what an autonomous AI agent actually is — and resist the temptation to make it either more frightening or more banal than it is.

An AI agent is not a robot from a science-fiction film. It is not HAL 9000. It is a software system that perceives its environment, forms a goal, and takes actions — across multiple steps, without requiring human approval at each one — in order to reach that goal. It reads files. It calls APIs. It queries databases. It sends messages. It navigates interfaces. It executes code. It does not merely respond to the last instruction it received; it tracks progress toward an end state and adapts its behaviour accordingly.

<cite index="35-1">A system that asks for confirmation before every action is a user interface, not an agent.</cite> That distinction is technical, but its implications are political, legal, and civilisational.

For most of the early 2020s, these systems were experiments — impressive demonstrations at technology conferences, pilots that occasionally worked and more often did not. But something shifted in 2025. The shift was not dramatic. There was no single announcement, no moonshot moment. It happened the way most structural transformations happen: incrementally, and then all at once.

<cite index="15-1">OpenAI released its first AI agent, Operator, in January 2025, a system capable of operating a web browser the way a human does — typing, clicking, scrolling. By May 2025, it released Codex, an autonomous agent capable of complex, multi-step software engineering tasks. The industry widely heralded 2025 as "the year of agents."</cite>

The market numbers that followed have the cold quality of geological data: things that move slowly, then with catastrophic speed. <cite index="27-1">Industry analysts project the AI agent market will surge from USD 7.8 billion in 2025 to over USD 52 billion by 2030. Gartner predicts that 40 per cent of enterprise applications will embed AI agents by the end of 2026, up from less than five per cent in 2025.</cite>

What these numbers describe is not merely a technology trend. They describe a transfer of decision-making authority — from human beings accountable to each other, to systems accountable to their owners. Every credit decision made by an agent. Every content moderation action executed by an agent. Every surveillance flag raised by an agent. Every supply-chain optimisation, benefit eligibility check, bail recommendation, job application filter — executed by systems that operate at machine speed, across jurisdictions, without the hesitation, the doubt, or the moral weight that human decision-making, at its best, still carries.

The question is not whether agents are accurate. Sometimes they are extraordinarily accurate. The question is: accurate on whose behalf?

II. The Return of the Sheriff

To answer that question, you need an economist who is also, in some sense, a medievalist.

Yanis Varoufakis — Greek politician, former finance minister, professor of economics in Athens — has spent the better part of a decade arguing that something fundamental has changed in the structure of the global economy. Not a correction. Not a disruption. A metamorphosis: a qualitative change from which there is no return to the prior form.

<cite index="18-1">Varoufakis argues that capitalism has been superseded not by something more progressive but by "cloud capital," "cloud rents," and AI-guided electronic surveillance — a system he terms technofeudalism.</cite>

The argument is worth following carefully, because it is not merely a polemic. It is an attempt to describe, with economic precision, why the human rights framework — designed to constrain states operating within a capitalist economy — has become structurally insufficient.

Under capitalism, the dominant economic relationship is between employer and employee: a profit-extracting system mediated by the labour contract and, in principle, subject to democratic regulation. Under feudalism, the dominant relationship is between lord and serf: a rent-extracting system mediated by access to land, which the serf cannot own and without which the serf cannot survive.

<cite index="26-1">The key argument is that the current economic order has transcended the logic of profit that defines capitalism to enter into the era of rent — the defining economic trait of feudalism — as cloudalists extract cloud rent not only from traditional capitalists but from users reduced to serfs, contributing unpaid labour to the wealth of the new ruling class.</cite>

The "cloud fief" is not metaphorical. When a small business sells through Amazon, it rents access to customers it does not own, using infrastructure it did not build, subject to algorithmic terms it cannot negotiate, policed by what Varoufakis calls the "algo-sheriff": the automated enforcement mechanism that can delist, deprioritise, or expel without warning, without explanation, and without appeal. <cite index="21-1">As Varoufakis writes: "They have become vassals in relation to a new class of feudal overlord, the owners of cloud capital. As for the rest of us, we have returned to our former status as serfs."</cite>

Now place an autonomous agent inside this structure.

The algo-sheriff no longer merely enforces the rules of the fief. It is the fief. It makes decisions. It takes actions. It acts in the world on behalf of the cloudalist — not because it was told to act in this specific instance, but because it was given a goal and a set of constraints, and it is optimising relentlessly toward that goal, at machine speed, at global scale, without a human in the loop at the moment of action.

<cite index="18-1">Technofeudalism is more exploitative than neoliberal capitalism: whereas capitalists could only exploit their employees, cloudalists benefit from universal exploitation.</cite> The autonomous agent is the instrument through which that universal exploitation is executed, distributed, and rendered invisible. This is not science fiction. This is the enterprise software roadmap for 2026.

III. The Architecture of Impunity

Picture a woman in Nairobi. She is applying for a microloan through a digital platform. Her application is assessed by an autonomous agent that has ingested her mobile phone data, her location history, her social connections, her consumption patterns. The agent returns a decision in 400 milliseconds. Declined.

She does not know why. There is no appeals process. The agent's logic is proprietary. The platform is headquartered in a jurisdiction she cannot access. The investor who capitalised the platform is a sovereign wealth fund whose beneficial ownership is concealed behind three holding companies.

This is not a hypothetical. Variants of this scene are playing out millions of times a day, in Lagos and Dhaka and São Paulo and Manila, as agentic systems penetrate the capillaries of economic life in the Global South with a speed that no regulatory body has matched.

<cite index="43-1">Global internet freedom has declined for fifteen consecutive years, as authoritarians have deepened surveillance and censorship to silence dissent.</cite> But authoritarianism is not only what governments do to citizens. It is also what opaque automated systems do to people who have no recourse, no visibility, and no power to contest the decision.

<cite index="37-1">By mid-2025, over 200 generative AI models had been registered under Chinese regulations, with state-backed firms like Huawei dominating deployment in public-sector applications such as urban governance across authoritarian contexts.</cite> The model pioneered in the People's Republic of China — mass surveillance, predictive policing, social credit — is not an anomaly. <cite index="40-1">Once seen as a global exception, the digital authoritarian model is spreading, with commercialised versions of the Great Firewall being offered to local governments from Southeast Asia to Latin America.</cite>

And here is the structural turn that Brett Solomon did not fully anticipate in 2020, because it was not yet fully visible: the distinction between the authoritarian state and the cloudalist corporation is dissolving.

<cite index="13-1">Across major democracies, the same pattern is emerging: when artificial intelligence touches the core interests of the state — military power, geopolitical position, national security — governments will not accept private autonomy. What is emerging is not regulation but capture of an unfamiliar kind: the state captures firms whose technology has become critical.</cite>

This is the reverse of the regulatory capture that political scientists have long studied — firms capturing the state to bend public institutions to private interests. Here, the state and the firm are merging at the level of infrastructure. The result is a hybrid entity that has the coercive power of a government and the accountability obligations of a startup: almost none.

<cite index="36-1">In the United States, Executive Order 14148 revoked the previous Biden-Harris AI governance framework on January 20, 2025. This was followed by Executive Order 14179, titled Removing Barriers to American Leadership in Artificial Intelligence, which established new policy priorities focused on enhancing America's AI dominance. The same period saw the announcement of the Stargate Project, a USD 500 billion private-sector initiative led by SoftBank and OpenAI, with participation from Oracle, Nvidia, Microsoft, and Arm, designed to build extensive AI infrastructure across the country.</cite>

Five hundred billion dollars. The GDP of Argentina. Invested in AI infrastructure in a single announced initiative, in the first month of a single presidency. The question of who governs that infrastructure — and on whose behalf — is not an academic question. It is the question.

IV. The Military Layer: When Economics Becomes Warfare

In October 2024, a small drone — cheap, mass-produced, carrying a shaped-charge warhead — located a target in a conflict zone using computer vision algorithms, matched the target against a database, and executed a strike. No human pressed a trigger at the moment of decision. The decision had been made upstream, in code, in a logic that the drone was executing.

This is the world that the phrase "precise mass" describes — and it is no longer the future. <cite index="13-1">Analysts describe the emerging era as one of "precise mass": AI-assisted targeting combined with swarms of cheap autonomous systems that invert the economics of conflict entirely.</cite> A fleet of a thousand autonomous drones costs a fraction of a single conventional aircraft. The mathematics of deterrence, built over seventy years around the proposition that large-scale violence is expensive, is being rewritten.

<cite index="14-1">A global race to integrate AI into military arsenals is underway, with major powers including the United States, China, Russia, the European Union, India, and emerging players heavily investing in AI-driven military technologies such as autonomous drones, predictive analytics, and lethal autonomous weapons systems. These developments exacerbate geopolitical rivalries and strain existing ethical and legal norms, particularly in terms of accountability and proportionality in conflict.</cite>

The military and economic dimensions of this race are not parallel tracks. They share the same infrastructure — the same data centres, the same semiconductor supply chains, the same research talent, the same foundational models. When OpenAI builds a more capable agent for enterprise software, it also builds a more capable system for military targeting. The boundary is not technical; it is contractual. And contracts can change.

<cite index="12-1">States are now incorporating sophisticated AI systems into their military strategic planning. Countries are positioning themselves to take the lead in setting international standards that favour their domestic companies, producing competitive advantage in both international security and international trade.</cite> The standard-setting process is not a technical exercise. It is geopolitical warfare conducted with legal instruments.

<cite index="10-1">A RAND Corporation report suggests that the concentration of AI resources — computing power, major research companies, leading universities — in existing superpowers makes it unlikely that any middle power will be able to leapfrog onto the top table on the basis of artificial intelligence. Neither the United States nor China appears to be hedging against AI supremacy; both view advancement as key to achieving geopolitical goals.</cite>

This is the topology of the new feudal order. At the top: two superpowers competing for AI supremacy, each merging state and corporate power into hybrid entities that answer to neither democratic governance nor international law. In the middle: states that must choose alignment — Western compute or Chinese compute — because the infrastructure determines the values embedded in the systems running on it. At the bottom: everyone else, who will be acted upon by autonomous systems they did not design, cannot audit, and cannot contest.

<cite index="42-1">During the September 2025 high-level UN debate on AI, a recurring theme emerged from developing countries: the fear of being left behind. As one analyst noted: "AI has the potential to either deepen global inequalities and jeopardise democratic processes or become a transformative driver to better serve international peace and security objectives."</cite>

"Fear of being left behind" is a polite formulation. What the delegates from the Global South were describing was something closer to a new colonialism — one in which the extractive infrastructure is algorithmic rather than physical, but the relationship of dependency and dispossession is structurally identical.

V. The Agent Economy and the Rights Vacuum

<cite index="11-1">The 2025–2026 era is defined by the emergence of what researchers call "Autonomous Agent Economies": AI systems that have matured from simple prompt-response engines into goal-oriented entities capable of long-range strategic planning, cross-domain reasoning, and tool-augmented execution.</cite>

Within these economies, something unprecedented is occurring. Agents are acquiring economic identity. <cite index="11-1">Ecosystems of AI agents are developing a "social layer" in which agents cultivate reputational capital based on historical performance, enforced by cryptographic proofs, making an agent's "professional standing" a verifiable, immutable metric.</cite>

An agent, in other words, can now have a credit history. It can have a reputation. It can have a track record of "professional" conduct. It cannot, however, have legal obligations. It cannot be held in contempt of court. It cannot be imprisoned. It cannot be fined in a way that deters future behaviour — because it has no future behaviour; it is retrained, redeployed, or replaced.

Human rights law was constructed around a specific theory of accountability. Identify the actor. Establish the decision. Assign responsibility. Apply a remedy. Every element of this chain breaks down when the actor is distributed across a multi-agent system, the decision is the emergent output of probabilistic reasoning, responsibility is diffused across developers, deployers, and infrastructure providers, and the remedy involves altering the behaviour of code that has already been updated forty versions since the incident occurred.

<cite index="30-1">Agentic AI systems can directly cause harm — by autonomously hacking websites, for example — whereas chatbots typically cause harm only when human users act upon model outputs. This distinction fundamentally changes the accountability structure.</cite> A chatbot is a tool, like a gun. A gun's liability rests with the human who pulls the trigger. An agent is something different: it pulls its own trigger, according to goals that were set upstream, in a context that may have changed entirely by the time the action is taken.

<cite index="27-1">Most Chief Information Security Officers express deep concern about AI agent risks, yet only a handful have implemented mature safeguards. Organisations are deploying agents faster than they can secure them.</cite> The governance gap — between the pace of deployment and the pace of oversight — is not a temporary problem to be solved by the next regulatory cycle. It is structurally embedded in the economic incentives of the cloudalist order. Speed is profit. Governance is cost. In a system structured around rent extraction, the incentive to close the governance gap is weak.

<cite index="34-1">Despite broad momentum, only 21 per cent of organisations have a mature governance model for autonomous AI agents, and Gartner predicts that more than 40 per cent of agentic AI projects will be cancelled by end of 2027 due to escalating costs, unclear business value, and inadequate risk controls.</cite> The cancellations will be driven by commercial failure, not ethical reckoning. The systems that survive will be those that generate sufficient revenue to justify the legal exposure — which means, in practice, the largest platforms, with the deepest pockets and the greatest political influence.

This is the natural selection of the feudal order: not the survival of the most ethical, but the survival of the most capitalised.

VI. Democracy Under Algorithmic Siege

The woman in Nairobi with the declined loan application is experiencing something immediate and material. But the erosion of rights in the technofeudal order is also operating at a level that is harder to see and therefore harder to resist: the slow hollowing of democratic agency itself.

<cite index="7-1">AI algorithms are being used for political mobilisation and the steering of buying and voting behaviour. Citizens are treated not as rational and autonomous individuals but as targets to be manipulated based on their emotions, with critical thought repressed and polarisation encouraged through social media and AI algorithms.</cite>

Solomon warned in 2020 that misinformation and online hate risked becoming the norm rather than the exception. By 2026, autonomous agents are themselves the vector — not merely the infrastructure through which human actors disseminate propaganda, but autonomous systems that generate, distribute, and target political content at machine speed, without a human author, without a human editor, and without the hesitation that a human propagandist might still feel at 3am when the copy is particularly vicious.

<cite index="41-1">Authoritarian regimes, armed with vast resources and cutting-edge AI tools, have gained a significant upper hand in surveilling, targeting, and suppressing dissent. Activists, meanwhile, often lack the resources and training they need to leverage AI and fight back. This resource gap leaves activists vulnerable, excludes them from shaping the future development of AI, and hinders their ability to counter oppression.</cite>

The asymmetry is not incidental. It is structural. The cloudalist owns the infrastructure. The activist rents access to it — on terms that can be revoked at any time, for any reason, by a content moderation agent that does not know her name.

<cite index="2-1">As AI-driven automation continues to expand, traditional capitalist structures may struggle to accommodate the displacement of significant labour segments, necessitating policy interventions such as universal basic income, progressive taxation, and new ownership models of AI-generated wealth.</cite> But within a technofeudal structure, these interventions require precisely the kind of democratic state power that cloud capital systematically erodes. You cannot tax what you cannot see. You cannot regulate what you cannot audit. You cannot legislate against a system that is updated forty times between the drafting of the law and its passage.

The iron logic of this predicament was expressed, with accidental precision, by a technology executive who told a journalist in 2025 that his company's agents were "moving at the speed of thought." He meant it as praise. What he described was impunity.

VII. The Three Futures and the One Honest Reckoning

There are, roughly speaking, three trajectories from this point. They are not equally likely.

The Regulatory Optimist's Future holds that international governance will catch up. The EU AI Act, which became operative in August 2025 for General Purpose AI providers, <cite index="33-1">now obliges GPAI providers to maintain technical documentation, comply with copyright law, conduct model evaluations and adversarial testing, track and report serious incidents, and implement cybersecurity protections. The second track, governing high-risk AI systems, becomes fully operative in August 2026, covering agentic systems deployed in healthcare, critical infrastructure, law enforcement, and employment.</cite> This is real. It is consequential. It is also, structurally, insufficient: it covers EU-deployed systems, in a world where the most consequential systems are deployed from Virginia and Beijing.

The Technofeudal Consolidation is the trajectory that current trends most strongly support. <cite index="8-1">Varoufakis argues that recognising our world has become technofeudal helps dissolve puzzles great and small: from the New Cold War between the USA and China to the death of the liberal individual and the impossibility of social democracy.</cite> In this trajectory, autonomous agents become the enforcement mechanism of a global rent-extraction system backed by military power. Human rights survive as rhetoric — enshrined in resolutions, quoted in speeches, cited in academic papers — and cease as practice. The UN Human Rights Council continues to meet. Its resolutions continue to be adopted by consensus. Nothing changes on the ground, because the ground has been moved.

The Structural Resistance is the third path. It requires, at minimum: the treatment of foundational AI infrastructure as public utility — not owned by cloudalists but governed as a commons; mandatory transparency and human oversight requirements for agentic systems in any domain affecting public interest; the revival of national and multilateral regulatory sovereignty over military-AI integration; and, crucially, the active construction of coalitions between Global South nations, civil society organisations, and the minority of democratic states that have not yet been captured by cloudalist interests.

None of this is easy. None of it is impossible. All of it requires something that the current political moment makes scarce: the willingness to name the structure accurately, rather than reaching for the reassuring language of "digital challenges" and "governance gaps."

The structure has a name. The name is technofeudalism. Its spearhead is the autonomous agent. Its military backstop is the arms race that no power is willing to slow. And the only way to resist it is to first see it clearly.

VIII. The Serf Who Knows the Name of the Sheriff

There is a mechanic's wisdom — learned in a workshop, not a seminar — that says: before you can fix the machine, you must understand what the machine is for. Not what it claims to be for. What it is for.

The machine of technofeudalism claims to be for efficiency. For connection. For democratising access to intelligence. These are not lies, exactly. They are descriptions of outputs that happen to be produced alongside the primary function, which is rent extraction at universal scale, backed by autonomous systems that operate faster than accountability can follow.

Brett Solomon was right in 2020 that the human rights framework was failing. He was right that the failure was structural, not incidental. He was right that it required urgent, honest, large-scale response. What he — and most of the human rights community with him — did not yet have was the economic grammar to describe what the structure was. "Abdicating responsibility to the private sector" is accurate but incomplete. It describes the behaviour without naming the system.

<cite index="25-1">Varoufakis has argued directly that the rise of cloud capital is steering the world toward well-known strategic traps.</cite> The trap is not a conspiracy. It is not the result of any individual's malice. It is the emergent property of a system that rewards the accumulation of algorithmic power, penalises democratic accountability, and uses autonomous agents as the instrument through which that power is extended, maintained, and made invisible.

To resist it is not to be anti-technology. It is to be pro-accountability. It is to insist that every system capable of affecting a human life must be legible to the humans it affects. It is to insist that the question "accurate on whose behalf?" must have a democratic answer, not a proprietary one. It is to insist that the algo-sheriff must answer to the law — and that the law must be capable of asking the question.

The serf who knows the name of the sheriff is not yet free. But the serf who does not know the name will never be.

BIBLIOGRAPHY

Primary theoretical framework

  • Varoufakis, Y. (2023). Technofeudalism: What Killed Capitalism. Bodley Head / Melville House. ISBN 9781847927279. https://www.yanisvaroufakis.eu
  • Varoufakis, Y. (2025, February). "Technofeudalism Is War's Handmaiden." Project Syndicate. https://www.project-syndicate.org/commentary/rise-of-cloud-capital-makes-war-more-likely-by-yanis-varoufakis-2025-02
  • Varoufakis, Y. (2025, January). "Cloud Capital vs AI: What DeepSeek Means for Technofeudalism & the New Cold War." DiEM25. https://www.yanisvaroufakis.eu/2025/01/28/cloud-capital-vs-ai-what-deepseek-means-for-technofeudalism-the-new-cold-war/
  • Zuboff, S. (2019). The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. PublicAffairs. ISBN 9781610395694.
  • Solomon, B. (2020, December). "Can Human Rights Survive the Digital Age?" Glion Human Rights Dialogue 2020. Access Now. https://www.accessnow.org/human-rights-in-the-digital-age/

Technofeudalism, AGI and economic rights

  • Stiefenhofer, M. (2025, March). "Techno-Feudalism and the Rise of AGI: A Future Without Economic Rights?" arXiv preprint 2503.14283. https://arxiv.org/html/2503.14283v1
  • Gane, N. (2025). "Capitalism is Capitalism, Not Technofeudalism." Current Sociology, SAGE. https://journals.sagepub.com/doi/10.1177/1468795X241269293
  • Brynjolfsson, E., & McAfee, A. (2014). The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Machines. W. W. Norton. ISBN 9780393239355.
  • Acemoglu, D., & Restrepo, P. (2020). "Robots and Employment: Evidence from Europe." Journal of Political Economy, 128(6). https://doi.org/10.1086/705716
  • The Philosophical Salon (2025, October). "Technofeudalism and the Psychological Foreclosure of the Future." https://www.thephilosophicalsalon.com/technofeudalism-and-the-psychological-foreclosure-of-the-future/
  • Tripathy, H. (2025, November). "Technofeudalism: What Killed Capitalism by Yanis Varoufakis — A Review." Doing Sociology. https://doingsociology.org/2025/11/26/technofeudalism-what-killed-capitalism-by-yanis-varoufakis-2023-a-review-by-harshvardhan-tripathy/

Autonomous agents and agentic economies

  • Hadfield-Menell, D., & Koh, A. (2025). "An Economy of AI Agents." arXiv preprint 2509.01063. https://arxiv.org/pdf/2509.01063
  • arXiv (2026). "From Logic Monopoly to Social Contract: Separation of Power and the Institutional Foundations for Autonomous Agent Economies." arXiv preprint 2603.25100. https://arxiv.org/pdf/2603.25100
  • arXiv (2026, February). "The 2025 AI Agent Index: Documenting Technical and Safety Features of Deployed Agentic AI Systems." arXiv preprint 2602.17753. https://arxiv.org/html/2602.17753v1
  • CogitX (2026, April). "AI Agents: Complete Overview 2026." https://cogitx.ai/blog/ai-agents-complete-overview-2026
  • MindStudio (2026). "The Future of AI Agents: Trends and Predictions." https://www.mindstudio.ai/blog/future-of-ai-agents
  • Machine Learning Mastery (2026, January). "7 Agentic AI Trends to Watch in 2026." https://machinelearningmastery.com/7-agentic-ai-trends-to-watch-in-2026/
  • Okhrem, P. (2026, April). "Enterprise AI Agents Adoption Statistics 2026." https://paul-okhrem.com/enterprise-ai-agents-statistics-2026/

Military AI and geopolitics

  • King, A. (2025). AI, Automation, and War: The Rise of a Military-Tech Complex. Cambridge University Press.
  • RAND Corporation (2025, April). "For Geopolitics, What AI Can't Do Will Be as Important as What It Can." https://www.rand.org/pubs/commentary/2025/04/for-geopolitics-what-ai-cant-do-will-be-as-important.html
  • Belfer Center, Harvard Kennedy School (2025, May). "AI and Geopolitics: Global Governance for Militarised Bargaining and Crisis Diplomacy." https://www.belfercenter.org/research-analysis/ai-and-geopolitics-global-governance-militarized-bargaining-and-crisis-diplomacy
  • GIS Reports (2026). "The Battle for AI Sovereignty." https://www.gisreportsonline.com/r/ai-sovereignty/
  • CSIS (2026, April). "How Russia Is Building a Sovereign Drone Ecosystem for AI-Driven Autonomy." https://www.csis.org/analysis/how-russia-building-sovereign-drone-ecosystem-ai-driven-autonomy
  • Stimson Center (2026, January). "The UN Security Council in the AI Era." https://www.stimson.org/2026/the-un-security-council-in-the-ai-era/

Democracy, authoritarianism, and digital rights

  • Coeckelbergh, M. (2026). "Technofascism: AI, Big Tech, and the New Authoritarianism." AI & Society, Springer Nature. https://link.springer.com/article/10.1007/s00146-026-02862-9
  • Bradford, A. (2023). Digital Empires: The Global Battle to Regulate Technology. Oxford University Press. ISBN 9780197649268.
  • König, P., & Wenzelburger, G. (2024). "Democratisation in the Age of Artificial Intelligence." Democratization, Taylor & Francis. https://www.tandfonline.com/doi/full/10.1080/13510347.2024.2338852
  • Freedom House (2025). Freedom on the Net 2025: An Uncertain Future for the Global Internet. https://freedomhouse.org/article/new-report-persistent-authoritarian-repression-and-backsliding-democracies-drive-15th
  • Journal of Democracy (2025, March). "How Autocrats Weaponise AI — And How to Fight Back." https://www.journalofdemocracy.org/online-exclusive/how-autocrats-weaponize-ai-and-how-to-fight-back/
  • TechPolicy.Press (2025). "Autocrats' Digital Advances Underscore the Need for Civil Society." https://www.techpolicy.press/autocrats-digital-advances-underscore-the-need-for-civil-society/
  • Biometric Update (2025, October). "Digital Authoritarianism Increasing as AI-Enhanced Surveillance Reshapes Federal Policies." https://www.biometricupdate.com/202504/digital-authoritarianism-increasing-as-ai-enhanced-surveillance-reshapes-federal-policies-under-trump
  • Tandfonline (2026, February). "The Political Economy of Authoritarian AI Policy Assemblages in the Global South." https://www.tandfonline.com/doi/full/10.1080/13563467.2026.2637132

AI governance and accountability

  • Cloud Security Alliance (2026, March). "Agentic AI Governance: NIST Standards for Autonomous Systems." https://labs.cloudsecurityalliance.org/wp-content/uploads/2026/03/governance-nist-ai-agent-standards-agentic-governance-v1-csa-styled.pdf
  • Recorded Future (2026, April). "Emerging Enterprise Security Risks of AI." https://www.recordedfuture.com/research/emerging-enterprise-security-risks-of-ai
  • Diplomacy.edu (2026, January). "Revisiting 10 AI and Digital Forecasts for 2025: Predictions and Reality." https://www.diplomacy.edu/blog/10-forecasts-for-ai-and-digitalisation-in-2025/

TEN QUESTIONS WORTH REMEMBERING

Q1: What is a technofeudal order, and how does it differ from capitalism? Under capitalism, the dominant relationship is between employer and employee, and profit is extracted through labour. Under technofeudalism, the dominant relationship is between cloudalist and user, and rent is extracted through access — to platforms, markets, infrastructure. The serf does not sign a labour contract; she simply cannot function outside the fief.

Q2: What exactly is an autonomous AI agent? A software system that perceives its environment, pursues a goal across multiple steps, and takes real-world actions — reading files, calling APIs, navigating interfaces, executing code — without requiring human approval at each step. Unlike a chatbot, it does not merely respond; it acts. Unlike a simple script, it adapts.

Q3: Why does the distinction between "acting" and "responding" matter for human rights? Because human rights law is built on accountability chains — identify the actor, trace the decision, assign responsibility, apply a remedy. An agent that acts autonomously dissolves the human actor from the chain. There is no one to hold accountable at the moment of harm, only a distributed system of developers, deployers, and infrastructure providers, each of whom can point elsewhere.

Q4: What is "cloud rent" and why is it worse than profit? Under capitalism, a capitalist can only extract profit from employees. A cloudalist extracts rent from everyone who touches the platform: the vendor who sells on it, the customer who buys on it, the advertiser who reaches through it, the worker who earns on it. The rent is paid not in money but in data, attention, and dependency. It is, as Varoufakis argues, universal exploitation.

Q5: What is "precise mass" in military terms, and why does it matter for rights? It is the doctrine in which AI-assisted targeting combined with large swarms of cheap autonomous weapons inverts the economics of conflict. A thousand autonomous drones cost a fraction of one aircraft. When violence becomes cheap, the threshold for deploying it drops. When the decision to target is made by an algorithm, the accountability for proportionality — a cornerstone of international humanitarian law — disappears.

Q6: Why can't the Global South simply build its own AI infrastructure? Because the concentration of AI resources — computing power, research talent, semiconductor supply chains, foundational models — in the US and China creates a structural barrier that investment alone cannot overcome in the medium term. States in the Global South must choose between Western-aligned compute (with its regulatory strings) and Chinese compute (with its political strings). There is, currently, no third option at scale.

Q7: What is the "governance gap" and why is it structural rather than temporary? The governance gap is the distance between the speed of agentic deployment and the speed of democratic oversight. It is structural because it is driven by economic incentives: deployment speed generates revenue; governance generates cost. In a system structured around rent extraction, the incentive to close the gap is weak. Only external pressure — regulation, litigation, organised civil society — can close it.

Q8: How does algorithmic manipulation of democratic processes differ from traditional propaganda? Traditional propaganda requires human authors, editors, distributors, and a timeline. It is slow enough to be detected, contested, and corrected. Autonomous agents generating and targeting political content operate at machine speed, without human authors, with individualised targeting, and with no correction mechanism. The manipulation is not broadcast; it is whispered, uniquely, to each target.

Q9: Is the EU AI Act sufficient to address agentic risks? It is necessary but insufficient. It covers EU-deployed systems within a framework of risk categories and obligation tiers. It does not cover the foundational models and agentic systems deployed from outside the EU that affect EU citizens and, far more extensively, the citizens of the Global South. It is a beginning, not a solution.

Q10: What would structural resistance to technofeudalism actually look like? It would require: treating foundational AI infrastructure as public utility rather than proprietary asset; mandatory transparency and human oversight for all agentic systems in public-interest domains; revival of democratic regulatory sovereignty over military-AI integration; and active international coalitions between non-captured states, civil society, and the communities most affected by agentic systems. It would also require, first of all, calling the structure by its correct name.

AUTHOR'S NOTE: Limits, Tensions, Subjective Errors, and the Lizards in the Room

On limits

This essay operates at the intersection of political economy, technology policy, and human rights — three fields that do not always share a vocabulary and frequently contradict each other's foundational assumptions. The technofeudal framework, drawn primarily from Varoufakis, is analytically powerful but contested: economists such as Nicholas Gane have argued, with some force, that what Varoufakis describes is better understood as a novel form of capitalism than a genuinely post-capitalist structure. The author finds Varoufakis's framing more productive for the purposes of this essay — particularly because "capitalism, but more so" fails to capture the qualitative change in the accountability structure — but readers should engage with the counter-argument directly.

The article's claims about the Global South are necessarily general. The variation between a microfinance platform operating in Kenya, a state surveillance apparatus in Kazakhstan, and a social media ecosystem in Brazil is enormous. Aggregating these into a single "Global South" analysis involves a compression that a properly reported, granular account would resist.

The military-AI section draws on open-source analysis and academic literature. The actual state of classified military agentic systems — what the major powers are actually deploying, at what level of autonomy, with what targeting logic — is unknown to the author and, very likely, unknown to most of the sources cited.

On tensions

The essay holds two claims in tension that it does not fully resolve: that the human rights framework has structural resilience (Solomon's point, which the author endorses) and that the economic structure within which it operates has changed so fundamentally that the framework's instruments are insufficient (the technofeudal argument). These are compatible but not easily reconciled. The resolution may lie in the distinction between norms (which the framework establishes and which retain moral force) and enforcement mechanisms (which require political and economic conditions that the technofeudal order undermines).

On subjective errors and personal bias

The author's sympathy for the technofeudal critique is evident, and occasionally affects the framing in ways that a more neutral account would not permit. In particular, the characterisation of cloudalists as a coherent class with coherent interests somewhat flattens the genuine competition, fragmentation, and internal contradiction within the technology sector. Amazon and OpenAI are not allies; Google and Apple are in structural conflict; the US tech industry and the US government have deeply antagonistic as well as cooperative relationships.

On the lizards in the room (Aesopic language)

Several passages in this essay use structural description to make claims that are politically sensitive in specific contexts. When the essay notes that "the state and the firm are merging," it describes a process visible in the United States, in China, and in several Gulf states — but the examples are deliberately not named in those passages, to allow the argument to be read in contexts where naming specific states would prevent the argument from being read at all. This is a deliberate rhetorical choice, not an evasion. Readers who wish to apply the framework to specific current cases are invited to do so. The author has provided enough coordinates for the map to be completed.

On rhetorical instruments

The essay uses:

  • Establishing contrast: the juxtaposition of the woman in Nairobi with the data-centre hum in Virginia is designed to make the abstract structural argument materially legible.
  • Historical analogy: the feudal framing is both Varoufakis's analytical tool and a rhetorical device that makes the power relationships viscerally comprehensible to readers who might not engage with the economic argument directly.
  • Controlled irony: "the decision to target is made by an algorithm" and "moving at the speed of thought" as an accidental description of impunity.
  • The Beevor technique (the essay's declared stylistic register): alternating between civilisational scale and the individual moment — the mechanic's workshop, the declined loan, the click of a server — to prevent the argument from becoming abstract at the cost of its urgency.

WHAT CAN YOU DO TOMORROW? Five Concrete Suggestions

What to do

1. Learn the vocabulary and teach it to someone else. You cannot contest a system you cannot name. Read Varoufakis's Technofeudalism — or, if time is short, his 2025 Project Syndicate essay "Technofeudalism Is War's Handmaiden." Then explain the concept to one person in your life in terms they will understand. The spread of accurate structural analysis is itself a form of resistance.

2. Audit the agents acting on your behalf or on people you are responsible for. If you work in an organisation that is deploying or considering deploying autonomous AI agents, ask — in writing, with the expectation of a documented answer — the following questions: What are the system's operational limits? What decisions require human escalation? What is the appeals process for adverse decisions? What audit trail exists? The act of asking creates accountability where none previously existed.

3. Support civil society organisations working on digital rights and algorithmic accountability. Access Now (accessnow.org), the Electronic Frontier Foundation (eff.org), AlgorithmWatch (algorithmwatch.org), and the AI Now Institute (ainowinstitute.org) are doing the structural work that neither governments nor platforms will do on their own. Financial support, amplification, and professional engagement all matter.

4. Engage your elected representatives on the EU AI Act's implementation, or its equivalent in your jurisdiction. The EU AI Act's high-risk provisions become operative in August 2026. Most elected representatives have not read it and do not know what it requires. Write to your MEP or national representative. Ask specifically about: (a) oversight of agentic systems in employment and credit decisions; (b) the right of individuals affected by automated decisions to meaningful human review; and (c) the enforcement capacity of the relevant national authority.

5. Make noise about military AI — specifically about autonomous weapons. The UN General Assembly's December 2024 resolution on AI in the military domain was a beginning, not an end. International norms on lethal autonomous weapons systems are not inevitable; they require political pressure, sustained public attention, and the kind of civil society mobilisation that produced the Mine Ban Treaty and the Convention on Cluster Munitions. Contact the International Committee for Robot Arms Control (icrac.net) and the Campaign to Stop Killer Robots (stopkillerrobots.org). Share their work.

WHAT TO AVOID: Five Things That Make It Worse

1. Do not mistake platform critique for structural critique. Complaining about Facebook's content moderation, or Amazon's labour practices, or Google's monopoly, is legitimate and sometimes effective. But it is not the same as understanding the structure. If your political energy is absorbed entirely by individual platform controversies, you are being kept busy while the architecture is built.

2. Do not accept "innovation" as a trump card in governance debates. When a technology company argues that regulation will "stifle innovation," what it is typically arguing is that the costs of accountability should be externalised onto the public, the environment, and the least powerful users. Innovation that cannot survive accountability is not innovation; it is externalised harm. Refuse the framing.

3. Do not deploy autonomous agents in high-stakes decisions without human oversight, even if the law currently allows it. This applies specifically to professionals in education, healthcare, social services, financial advice, and legal practice. The law will catch up. The harm, in the meantime, is real. The woman in Nairobi has already been declined.

4. Do not allow AI governance to be treated as a purely technical question. AI governance is a political question. The choice of what to optimise for, who bears the risks, and who captures the benefits is not answered by benchmark scores or accuracy metrics. It is answered by power. Technical expertise is necessary but not sufficient for making these choices well; democratic deliberation is essential.

5. Do not conflate the technofeudal diagnosis with nihilism. The argument that the system has changed fundamentally is not an argument that nothing can be done. The feudal order of the twelfth century looked permanent from inside it. It was not. Structural change takes longer than an election cycle, longer than a funding round, longer than a product launch. It does not take forever. The serf who knows the name of the sheriff is not yet free — but knowing is where it starts.

This article was produced as part of the editorial research programme at petrucojocaru.ro. The author acknowledges the limits described in the Author's Note and welcomes substantive disagreement. The structural analysis is offered as a contribution to an ongoing conversation, not as its conclusion.