Blog · Review Essay · Last reviewed June 19, 2026

A Hacker Manifesto and the Vectoralist Class

McKenzie Wark's A Hacker Manifesto is a compact theory of information-age class power. It matters now because AI platforms have made the book's central conflict concrete: people and machines produce abstractions, but the owners of the vectors decide which abstractions can circulate, be priced, be trained on, be queried, and be turned into institutional authority.

A vector, in this review, is any controlled route that determines whether an abstraction can be created, discovered, invoked, trusted, priced, audited, or contested. Vectoral power exists where control over that route matters more than formal ownership of the artifact moving through it.

The AI-era test is practical: if a creator, worker, agency, school, lab, or user cannot move the model, data, metadata, logs, identity, payment, or appeal record without permission from the channel owner, the channel is part of the power relation.

The Book

A Hacker Manifesto was published by Harvard University Press in 2004. Google Books lists the edition at 208 pages, published October 4, 2004, with ISBN 9780674015432 and subjects including computers, internet, intellectual property, and political philosophy. The same record describes the book as a restatement of Marxist thought for cyberspace and globalization, drawing on Guy Debord and Gilles Deleuze, and centered on the conflict around commodified information.

Wark's own institutional location matters. The New School profile lists A Hacker Manifesto among Wark's publications and names media theory, new media, and critical theory as research interests. This is not a manual about intrusion or security technique. It is media theory written as political economy: a manifesto about who produces new abstractions and who owns the channels through which those abstractions become valuable.

The book belongs beside Hackers, Coding Freedom, Protocol, The People's Platform, Data Cartels, and The Stack. Those books trace hacker culture, open-source institutions, network control, platform dependency, information monopoly, and planetary computation. Wark supplies the class vocabulary: hackers make the new; vectoralists own the vectors.

Current Context

As of this review date, the vectoralist question has moved from cultural theory into infrastructure governance. The European Commission's Digital Markets Act materials describe gatekeepers as large platforms that provide core services such as search engines, app stores, and messaging services. Its Data Act materials add a different pressure point: access to connected-device data, cloud switching, data portability, and constraints on unfair contracts. These are not Wark's categories, but they show regulators now treating routes, not only artifacts, as governable objects.

AI makes that shift sharper. The Open Source AI Definition 1.0 distinguishes meaningful open-source AI from mere access to weights by requiring practical freedom to use, study, modify, and share, supported by the preferred form for modification: data information, code, and parameters. The EU General-Purpose AI Code of Practice, published July 10, 2025, turns transparency, copyright, and safety and security into documentation work for AI Act compliance. NIST's 2026 AI Agent Standards Initiative treats agent identity, interoperability, and secure action on behalf of users as standards problems. OMB's 2025 federal AI acquisition guidance warns agencies to pay attention to vendor sourcing, data portability, interoperability, data rights, privacy, documentation, and lock-in when buying AI.

That current context refines Wark's vocabulary. A vector is not only a telecommunications line or a media platform. It is the bundle of routes that decides whether an abstraction can become operational: training-data access, model weights, inference gateways, identity credentials, app stores, agent permissions, payment rails, provenance metadata, audit logs, procurement terms, and exit rights.

The Hacker Is Not Just a Programmer

The first useful move in A Hacker Manifesto is widening the word "hacker." Wark does not limit the hacker class to computer programmers. A hacker is anyone whose labor produces new information, new relations, new concepts, new expressions, new forms, or new abstractions from raw material. That includes software developers, scientists, artists, writers, designers, musicians, theorists, biologists, and other makers of information-bearing novelty.

This broad definition can irritate readers expecting a narrower history of computer hacking. A First Monday review from 2005 notes that Wark's hacker includes writers and singers alongside programmers, which is exactly where some readers may feel the book leaving conventional hacker culture. Brent Jesiek's 2006 New Media & Society review makes a similar point from a more sympathetic angle: the book is not a conventional account of hackers, but an information-age Marxism about a new productive class and the class that controls informational conduits.

The broadness is the point. Wark is not trying to describe a subculture. Wark is trying to name a mode of production in which valuable work increasingly appears as abstraction. Code is an abstraction. A model architecture is an abstraction. A dataset schema is an abstraction. A design pattern, scientific paper, video format, patent claim, playlist genre, legal database, prompt template, benchmark, and synthetic voice are all abstractions that can be copied, priced, enclosed, ranked, and routed.

That makes the hacker class vulnerable in a specific way. If the output of labor is information, then the central fight is not only wages. It is ownership, licensing, distribution, access, credit, versioning, search visibility, API permission, and the right to keep building without asking permission from whoever owns the channel.

The Vectoralist Class

Wark's strongest term is "vectoralist." The vectoralist class owns and controls the vectors through which information moves: networks, platforms, databases, distribution channels, legal rights, telecommunications systems, search systems, archives, standards, payment rails, identity layers, and other conduits that turn abstraction into power.

For AI, that definition has to be made operational. A vector is any controlled path that determines whether an abstraction can be made, moved, invoked, ranked, monetized, trusted, or acted on. Compute clusters, cloud accounts, model hubs, app stores, connector directories, package registries, benchmark leaderboards, search indexes, copyright licenses, provenance metadata, enterprise identity systems, payment rails, and agent logs are all vectoral surfaces. They are not just infrastructure around intelligence. They decide which intelligence becomes usable.

Four questions keep the term disciplined. Access: who can reach the route, and on what terms? Translation: what metadata, ranking, format, schema, benchmark, or policy turns an abstraction into something usable? Authority: who can invoke the route to act on another person, record, market, or institution? Exit: can the user, worker, creator, or buyer leave with enough data, logs, code, and rights to keep functioning elsewhere?

The term has aged well because it does not depend on one company, device, or business model. In 2004, the immediate setting was file sharing, software, intellectual property, digital culture, and globalization. In 2013, Melissa Gregg's Los Angeles Review of Books interview with Wark framed the book's ten-year legacy around the shift from factory-centered class analysis to a knowledge economy in which hackers produce innovation, knowledge, and abstraction while vectoralists appropriate and commoditize those goods. The interview also captured Wark's later judgment that social creation had won some affordances while vectoral power regrouped around metadata and more abstract control.

That later comment is crucial. The vector is not only the visible file, post, song, model, article, or app. It is also the metadata, the identity graph, the query log, the ranking signal, the dependency map, the training corpus, the access token, the permission scope, the benchmark result, the provenance field, the container image, the distribution agreement, and the payment event. The owner of the vector can make openness feel generous while keeping control over the layer that makes openness operational.

This is why the book remains sharper than generic "information wants to be free" rhetoric. Wark is not saying that copying alone defeats property. Wark is saying that information creates a distinct class struggle because it can be nonrival at the level of the artifact and still scarce at the level of access, attention, legality, computation, discoverability, and institutional adoption.

The AI Reading

Read in 2026, A Hacker Manifesto looks like a prehistory of AI platform politics. Foundation models are machines for producing and manipulating abstraction, but they sit on vectors owned by firms and institutions: training data pipelines, cloud compute, GPU supply chains, model weights, API gateways, app stores, eval suites, safety classifiers, enterprise connectors, identity systems, payment rails, and default interfaces.

The hacker class now includes more than people writing code or text. It includes open-source maintainers whose repositories become training material, artists whose styles become latent coordinates, scientists whose papers become retrieval inputs, forum users whose answers become model behavior, moderators whose judgment becomes safety data, labelers whose decisions become alignment examples, and users whose prompts become telemetry. The work of abstraction is distributed. The ownership of the vector is concentrated.

The current policy record makes the vectoral question more than a metaphor. The European Commission describes Digital Markets Act gatekeepers as large platforms that provide core services such as search engines, app stores, and messaging services. The Open Source Initiative's Open Source AI Definition 1.0 says open-source AI requires freedom to use, study, modify, and share, and it treats the preferred form for modification as including data information, code, and parameters. The EU's General-Purpose AI Code of Practice, published July 10, 2025, organizes AI Act compliance around transparency, copyright, and safety and security. NIST's February 2026 AI Agent Standards Initiative names autonomous agents, secure action on behalf of users, identity, security, and interoperability as standards problems. These are different regimes, but they all circle the same terrain: who controls the path from information to action?

That is not a simple anti-AI argument. Models can help hackers make more abstractions: code, images, interfaces, simulations, theories, diagrams, datasets, tests, and tools. But Wark's class lens asks who captures the surplus when abstraction becomes cheap to produce. If every worker gains a copilot while every platform gains the telemetry, distribution channel, billing relationship, and model improvement loop, the balance of power may still move upward.

The same issue appears in open models. A model weight release can expand technical agency, but openness at one layer does not settle vectoral control at other layers. The OSI definition is useful precisely because it refuses to treat weights alone as the whole system: meaningful modification also depends on training-data information, code, and parameters. Even then, practical power still depends on compute, hosting, package registries, model hubs, benchmark leaderboards, safety certification, deployment policy, and enterprise procurement paths. The vector can absorb the gift.

The intellectual-property fight is another vectoral fight. The U.S. Copyright Office's AI initiative received more than 10,000 comments in response to its 2023 Notice of Inquiry and released a pre-publication Part 3 report on generative-AI training on May 9, 2025, with the final version expected without substantive changes to the analysis or conclusions. That does not settle training disputes, but it shows that the conflict Wark named has moved into law: the abstraction factory now depends on contested claims about what can be copied, learned from, licensed, summarized, opted out of, and monetized.

Wark's vocabulary is useful because it avoids treating "open" and "closed" as moral absolutes. The real question is what practical powers move with the artifact. Can affected people inspect it, run it, repair it, fork it, contest it, exit it, and use it without feeding a dependency loop they cannot govern?

Recursive Reality

The book also helps explain recursive reality: representations become infrastructure, then infrastructure changes the world being represented.

A search index ranks documents, so publishers write for the index. A recommender ranks videos, so creators make videos for the recommender. A benchmark ranks models, so labs train toward the benchmark. A coding assistant suggests patterns, so codebases absorb the assistant's defaults. A legal research platform organizes precedent, so lawyers learn the platform's map of relevance. A model marketplace ranks downloads, so builders shape releases around distribution signals. The vector does not merely transmit abstraction. It trains the next abstraction.

This is where A Hacker Manifesto becomes more than a property argument. It is a theory of world-making through information channels. Once the vectoral class owns the path by which abstractions reach users, it can shape not only prices but imagination. Some tools become visible. Some problems become profitable. Some forms of knowledge become easier to cite, retrieve, train, and operationalize. Others become noise.

AI systems intensify the loop because they can act on the abstractions they inherit. A model trained on platform-shaped culture produces new platform-shaped culture. A procurement benchmark produces products optimized for procurement benchmarks. A data broker's schema becomes the memory through which agencies see people. A school dashboard produces student records that train future dashboards. The vector observes the world, routes the world, and then treats the routed world as evidence.

Governance and Safety

The governance lesson is to attach duties to vectors, not only to artifacts. A model card, open-weight release, copyright notice, or benchmark score is weak if the surrounding route remains uninspectable. The safety unit is the action route: data enters, a model transforms it, a router selects a provider or policy, an identity system grants authority, a tool executes, and logs or provenance records make later reconstruction possible. If one of those route layers is hidden, artifact-level transparency can collapse into theater.

The relevant duties are layer-specific: data documentation and consent or licensing analysis for corpora; provenance and watermarking where they are reliable enough to help; compute and hosting disclosure for material dependencies; routing and version records for gateways; identity, authorization, and revocation controls for agents; appeal paths for ranking, takedown, suspension, or refusal; and procurement terms that preserve export, deletion, logs, and exit.

A practical vector audit should therefore map six records before deployment: the artifact record, the route record, the authority record, the compensation or billing record, the incident record, and the exit record. For an AI agent, that means not only a model card but a tool-permission map, service-account identity, action log, approval policy, rollback path, and evidence-retention rule. For a model marketplace or gateway, it means provider selection criteria, fallback behavior, model-version identity, region controls, cache status, and pricing incentives. For creative or scholarly systems, it means training-data documentation, provenance limits, opt-out handling, licensing assumptions, and routes for correction or dispute.

Safety is also layer-specific. Open weights can support audit, repair, local control, and scientific study, but they can also make misuse harder to contain. Closed APIs can support monitoring and abuse response, but they can also centralize telemetry, pricing power, unilateral policy changes, and public dependence on private chokepoints. Wark's vocabulary helps avoid both easy slogans: openness is not automatically liberation, and enclosure is not automatically safety.

The current policy vocabulary partly matches this frame. The Digital Markets Act treats certain core platform services as gatekeeper routes. The Data Act puts access, portability, cloud switching, and unfair contract terms on the same infrastructural map. The Open Source AI Definition 1.0 asks whether users have practical freedom to use, study, modify, and share, including access to data information, code, and parameters. The EU General-Purpose AI Code of Practice turns transparency, copyright, and safety into documentation work for AI Act compliance, including Article 53 obligations for general-purpose AI model providers. NIST's 2026 AI Agent Standards Initiative adds the next vector: interoperable, secure agents that can act through user credentials and tools. The U.S. Copyright Office's AI reports add another layer by showing how training-data disputes become infrastructure questions, not only author-rights disputes.

Those instruments do not solve vectoral power by themselves. They are useful because they move oversight toward the route: who can see the corpus, change the model, operate the gateway, authorize the agent, inspect the log, appeal the decision, and leave without losing institutional memory.

Where the Book Needs Friction

A Hacker Manifesto is brilliant as a conceptual machine, but it is not a careful empirical map of every labor relation in the information economy. Its manifesto form is compressed, polemical, and aphoristic. That makes the book memorable. It also makes it easy to overextend.

The biggest risk is class flattening. Wark's hacker class includes artists, scientists, programmers, writers, and other abstraction workers, but those groups do not share one stable interest. Some are precarious. Some are employees of vectoralist firms. Some become founders, investors, managers, or owners. Some depend on copyright or patents for survival. Some produce public knowledge; some produce surveillance systems. Some want openness; some want monopoly if they can get it.

Jesiek's New Media & Society review is useful here because it praises the ambition while noting concerns about the likely unification of productive classes and about making class primary over other relations of oppression. That caution has only become more important. AI extraction, data labor, biometric surveillance, platform moderation, algorithmic ranking, and intellectual-property fights are also shaped by race, gender, disability, migration status, geography, language, caste, credentialing, and the uneven capacity to refuse.

The book also has an ambivalent relationship to institutions. A politics of the commons cannot live on circulation alone. Public-interest infrastructure needs libraries, universities, standards bodies, unions, maintainers, courts, archives, regulators, public compute, cooperative platforms, durable funding, and maintenance cultures. Without institutions, the vectoralist class often wins by being the only actor willing to keep the channel running.

Those limits do not weaken the book's relevance. They make its best use clearer. Read it as a diagnostic for information power, not as a complete program. It names a conflict that has become more visible in the AI era: abstraction is socially produced, but the route from abstraction to reality is owned.

What This Changes

The practical lesson is to audit the vector, not only the artifact.

When evaluating an AI model, platform, agent, dataset, creative tool, search product, or open-source release, ask who owns the route. Who controls discovery, hosting, identity, billing, logging, metadata, evaluation, compliance, and defaults? Who gets to change the terms after people depend on the system? Who receives the improvement data? Who can fork in practice, not just in license theory? Who can appeal when the vector misclassifies, buries, prices, blocks, or extracts from them?

For governance, split the vector into layers before deciding what "openness" or "safety" means. Compute decides who can train, test, fine-tune, and audit. Data decides whose work becomes substrate and under what consent or license. Model weights decide who can run or inspect a system. Model routing decides which system answers. App stores and agent directories decide which capabilities become available. Identity and payment rails decide who may act and transact. Logs and provenance decide what can be reconstructed after harm. Procurement decides whether any of those promises survive renewal, price changes, vendor substitution, or exit. A serious assessment has to map all of those layers, not only the visible output.

For labor, ask whether AI expands the hacker's agency or merely accelerates abstraction for someone else's channel. A coding assistant can improve craft if it leaves workers with understanding, authorship, bargaining power, and repair capacity. It can also turn craft into prompt throughput while centralizing knowledge in a vendor's model and logs.

For culture, ask whether new creative abundance is paired with durable rights for creators, maintainers, labelers, communities, and publics. A platform that celebrates remix while owning ranking, monetization, training access, and moderation has not abolished property. It has moved property into the vector.

That makes the review a companion to agent-store governance, open-weight model politics, release-boundary questions, model routing, compute governance, AI procurement, content provenance, data enrichment labor, platform governance, platform monopoly power, and search remedies. A Hacker Manifesto matters because it gives a hard name to a soft interface. The friendly surface says create, share, prompt, remix, build, publish, and connect. The vector decides what can be found, paid, trained, trusted, remembered, and appealed. In the AI era, that is where much of the politics lives.

Source Discipline

This review separates Wark's manifesto vocabulary from current governance evidence. Harvard University Press, Google Books, and The New School support book and author facts. Reviews and interviews help interpret the hacker/vectoralist vocabulary but do not prove every AI-era extension. OSI, European Commission, NIST, OMB, and Copyright Office materials support the current policy context; they are not treated as evidence that any one legal regime has solved vectoral power.

Claims about open-source AI, copyright, data portability, procurement, and agent standards are jurisdiction- and instrument-specific. The OSI definition is a community standard, the EU GPAI Code is a voluntary compliance tool connected to the AI Act, the Digital Markets Act and Data Act are EU instruments with their own scopes and enforcement paths, NIST standards work is not a statute, OMB memoranda govern federal agency use and acquisition, and the Copyright Office report is policy analysis rather than final case law. Keeping those categories separate is part of the point: vectoral power often grows when legal, technical, and cultural claims blur together.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books