Who Owns the Future You?

When Context Becomes the Primary Means of Production

Begin with a strong assumption: artificial superintelligence arrives within the next five to ten years; human intelligence becomes far weaker than AI intelligence; and 99% of a person's value at work comes from a long-accumulated personal knowledge base that AI can call as context.

This is not a prediction about the ASI timetable, the relative abilities of humans and AI, or the literal accuracy of the 99% figure. It is a thought experiment. Its purpose is to make one institutional question impossible to avoid: if general intelligence becomes a machine that anyone can rent, does context become a person's most important means of production? If it does, when should that context belong to a company, and when should it remain with the individual?

The central conclusion is that context will become a primary means of production, but it cannot coherently be awarded as one indivisible object either to the person or to the company.

A company may control work facts, trade secrets, proprietary bindings, and expressly commissioned deliverables. An individual must retain an inalienable measure of cognitive continuity and general capability. Value jointly created by a person, a team, and company resources requires shared governance. Data about customers and colleagues can only be held in trust for legitimate purposes; it is not the absolute property of either side.

In plainer terms, a worker may carry away the method, but not the customer. A company may keep the process, but it should not permanently and freely possess a former employee's digital double.

The apparent ownership question therefore needs to be decomposed into at least six separately configurable rights. Access and continuity asks who may keep consulting and calling the context after an exit, acquisition, or system migration. Use and execution asks who may let AI reason from it, invoke tools, or act externally, for which purpose and under which authority. Modification and training asks who may correct, aggregate, transform, or use it to train an embedding, adapter, agent, or other derived model. Export and transfer asks who may carry away an original, a sanitized representation, or a contribution attestation, and who may license or sell it to a third party. Exclusion and deletion asks who may reject a new purpose, stop identity replication, revoke credentials, or require a version and its derivatives to be deleted. Audit and revenue asks who may inspect use, verify provenance, allocate risk, and participate in a license fee, buyout, royalty, or collective dividend.

These operating rights sit on top of distinct legal and institutional interests: custody of source facts, ownership of fixed expression and work product, cognitive continuity in general methods, duties of confidentiality and privacy, limits on training and continued deployment, and governance of jointly produced value. The decomposition matters because the right to retain a document need not imply a right to train a digital double, and a right to use a method need not imply a right to export its secret inputs.

These rights can point in different directions at the same time. A company can own a particular document without owning the underlying general method. A worker can retain a method without being entitled to disclose the secret inputs through which it was learned. A team can create an important workflow without any single member being entitled to sell all of it. A customer can have privacy interests in information that neither the worker nor the employer may treat as freely alienable property. The question is therefore not simply, "Who owns the database?" It is, "Which rights attach to which layer of context, for which purpose, for how long, and under what accountability?"

When Context Becomes Part of the Person

Andy Clark and David Chalmers' theory of the extended mind proposes that an external system can become part of a cognitive system when it participates in memory and judgment in a stable, trusted, and readily available way. Under the ASI assumption, a personal knowledge base may no longer be an ordinary collection of files. It may operate as an extension of personal memory, a continuous carrier of professional judgment, and a cognitive organ through which a person sets goals, directs AI, and remains responsible for decisions. Disconnecting it may therefore remove not merely an archive but the conditions under which the person can continue to think and work in a characteristic way.

Margaret Jane Radin's theory of property and personhood strengthens this argument. Some external resources become so bound up with the development and expression of personhood that treating them as ordinary, fully alienable commodities is morally inadequate. Hegel's distinction between transferable products of particular labor and a person's inalienable personality and general capacities points in the same direction. Kant's demand that people not be treated merely as tools reinforces the limit. A contract may assign defined deliverables, limited services, and company-specific implementations to the firm, but standard-form employment terms should not permanently transfer a worker's entire professional memory, general judgment function, and future capacity to act merely because the worker clicked "accept" or created a record on an employer's device.

Together, these ideas support a mandatory floor: a right to cognitive continuity. Even when a person may not take the company's raw material, the person should be entitled to leave with a representation that does not disclose protected information but is functionally sufficient to continue their general capacity for judgment. This might take the form of a sanitized skill core, a portable account of methods, or an attestation of demonstrated capabilities. It is not necessarily a copy of the underlying documents, and the right is a normative proposal rather than a generally enacted legal right.

This right matters because a future knowledge base may do more than remind a worker of facts. It may encode how that worker recognizes a pattern, frames uncertainty, sequences a diagnosis, or knows which question to ask. If all of that can be confiscated at the boundary of employment, resignation begins to resemble professional amnesia. The employer would not merely own the output of past labor; it would control the infrastructure required for the worker's future labor.

Locke's labor argument supplies only half an answer. A worker creates new value by selecting, labeling, reflecting on, and structuring material, so the worker has a moral claim to the added structure. But mixing labor with information cannot convert pre-existing company secrets, customer data, or colleagues' contributions into the worker's personal property. The reverse inference also fails: providing wages and servers does not entitle the company to every general capacity that the worker develops during employment. Because several parties make contributions that cannot be completely specified in advance, neither "who owned the input" nor "who paid the wage" resolves every right in a composite context.

Republican freedom adds the problem of non-domination. Freedom does not mean only that another party happens not to interfere today; it requires that the other party lack arbitrary power over the basic conditions of action. The Stanford Encyclopedia of Philosophy discussion of republicanism is directly relevant: if an employer can block 99% of a person's productive capacity at will, the worker remains structurally dependent even if the employer never presses a delete button. "You can quit" is not a genuine exit when quitting destroys the means of working. A genuine exit requires semantic portability, a transition period, rights to challenge and audit, and a portable core that allows continued work without disclosing protected facts.

Yet cognitive integration does not erase other people's rights. "I lived through it" does not mean "I am authorized to disclose it." A person may remember a customer's quoted price, a colleague's illness, an unreleased product, or an internal threshold. The fact that these memories affected the person's judgment does not transform the facts into personal property. The theory of contextual integrity makes the limit clear: an appropriate information flow depends on roles, information types, recipients, and transmission norms, not simply on whose mind or notebook once contained the information. A right to continuity is therefore a right to a non-infringing functional representation, not a license to carry away confidential, personal, or regulated source material.

Team context imposes a second limit. Social knowledge and transactive memory—who knows what, how each person judges, and how members correct one another—are not the independent product of any one participant. A star employee should not privately enclose them, but neither should they become the absolute property of shareholders merely because the company controls the server. An Ostrom-style bundle of rights is a better model: access, management, exclusion, transfer, and revenue can be assigned separately instead of granting one unlimited owner every power.

The philosophical boundary is consequently dual. Context can become sufficiently integrated with a person that complete forfeiture would injure autonomy and personhood. At the same time, the context remains relational: it may contain the lives, promises, investments, and legitimate secrets of others. A defensible institution must preserve the person without appropriating the factual world of the company, the team, the customer, or the colleague.

There Is No Single Legal Ownership of Context

Current law does not generally recognize a unified object called "context" with one owner. In the United States, California, and the European Union, a knowledge base would instead be disassembled into different legal objects, interests, duties, and remedies. The discussion here is structural analysis, not legal advice for a particular contract, jurisdiction, person, or dispute.

Documents, code, and the particular expression of a company-specific skill file created within the scope of employment may often be treated as work made for hire, with copyright initially held by the employer under 17 U.S.C. Section 201(b). That does not mean the employer receives a copyright monopoly over ideas, procedures, processes, systems, or methods of operation. 17 U.S.C. Section 102(b) separates protectable expression from those uncopyrightable methods. Owning the document's copyright does not, by itself, privatize every capability described by the document.

The absence of copyright protection does not make a method freely portable. A process, model, customer list, pricing rule, or operational method can qualify as a trade secret under 18 U.S.C. Section 1839 when it derives value from not being generally known and is subject to reasonable efforts to preserve secrecy. California decisions likewise distinguish a former employee's general knowledge, skill, and experience, which the employee may use, from trade secrets and confidential customer information, which the employee may not use or disclose. Morlife, Inc. v. Perry shows how customer information can cross that boundary.

The EU Trade Secrets Directive 2016/943 similarly states that trade-secret protection should not restrict employees' use of experience and skills honestly acquired in the normal course of employment or be used to impose unwarranted restraints on labor mobility. The point is not that all methods belong to workers, but that secrecy rules must distinguish protected business material from general professional capacity.

The boundary for software and inventions can also depend on specific state law. California Labor Code Section 2870 gives limited protection to an invention developed entirely on personal time without employer resources. That protection can narrow when the invention relates to the employer's business or anticipated research, or results from work performed for the employer. The computer on which an artifact was created is therefore one evidentiary factor, not a universal switch that assigns every right in the surrounding context.

Data-protection rights add another layer but do not settle ownership. Access, correction, deletion, or portability rights under the GDPR or the CCPA are procedural rights concerning personal data. They do not give a worker title to the employer's entire knowledge base, nor do they automatically override the privacy rights of colleagues and customers, trade-secret protections, retention duties, or other lawful limits. The EDPB portability guidance generally distinguishes data actively provided by a person and data observed from that person's activity from inferences or derived data generated through the controller's analysis. Customer, colleague, and supplier records also implicate other people's rights; the company may itself be a purpose-bound controller or steward rather than an absolute owner.

Derived artifacts do not automatically escape these constraints. Turning protected source material into an embedding, summary, adapter, behavioral profile, knowledge graph, or model weight does not "wash away" the underlying secrecy, privacy, copyright, contractual, or licensing restrictions. A representation that enables reliable reconstruction or extraction of protected facts remains risky even if its file format is novel. The relevant inquiry includes what the artifact reveals, reproduces, or permits a model to do, not merely whether it contains verbatim text. Conversely, fixing a worker's general judgment in model weights should not make that judgment permanently free to the company merely because its format changed.

The rough present-day baseline is therefore this: the company often controls fixed work product and protected business secrets, while the individual retains general knowledge, skill, and experience. The emerging legal rupture is that abilities once held only in a worker's mind are now fixed in prompts, skill files, behavioral traces, fine-tuning records, memories, adapters, and model weights. Once fixed, they can be pulled toward the company side by existing rules for work product, confidential information, trade secrets, and broad contract assignments, even when they embody capabilities that previously would have followed the worker as general skill.

Current law has not produced stable answers to several resulting questions. May a company run a worker's judgment model forever after departure? May the worker demand a representation that removes company facts while preserving general capability? How should revenue from a jointly produced adapter, embedding, or agent memory be allocated? Is consent meaningful when continuous behavioral monitoring occurs inside an unequal employment relationship? Can a personal cognitive profile be sold with the assets when a company is acquired or becomes insolvent? International Labour Organization research on workers' data rights identifies the gaps created by structural power asymmetry. The EU Platform Work Directive 2024/2831 supplies precedents in algorithmic transparency, human review, and worker-representative participation, but it is not a unified law of context ownership.

That rupture cannot be solved by calling every artifact "company data" or, conversely, by labeling every learned capability "personal memory." It requires attention to source, expression, inferential power, purpose, identity replication, and the rights of third parties. It also requires contract terms that separately address AI training and post-employment model use rather than hiding those permissions inside a general assignment of work product.

Why Both Absolute Ownership Regimes Fail

If the company owns all context, leaving a job becomes a form of professional forgetting. The employer becomes the sole buyer of the worker's productive capacity because only the employer controls the context needed to activate it. Workers who know that careful documentation will train an AI to replace them will rationally record less, share less, and provide less training. The company pays a salary once but may acquire a worker's judgment model, capable of being copied and operated indefinitely at near-zero marginal cost.

This is cognitive enclosure. The worker is separated not only from the product of labor but also from the means of production required for future labor. The enclosure is especially severe when the system preserves recognizable judgment, style, or identity after the person leaves. Under the 99% assumption, the enclosure is not a routine file-access rule; it is structural control over the worker's future productive personality.

Becker's human-capital analysis distinguishes general investments, which create value across firms, from firm-specific investments, whose value depends on a particular organization. An ASI-era institution should translate that distinction into a portable skill core and a nonportable company binding rather than forcing one owner to take the entire file. Grossman and Hart's incomplete-contract theory treats ownership as residual control over matters a contract did not specify. Giving all residual control to one party weakens the other party's incentive to invest. Workers invest in judgment, relationships, and reflection; firms invest in data, colleagues, brands, risk, and infrastructure. A durable arrangement must leave each side enough control to keep investing.

Polanyi's The Tacit Dimension begins from the proposition that people know more than they can say. One economic function of advanced AI is to encode, copy, and execute tacit judgment that was previously difficult to transfer. Lower copying costs do not decide who deserves the gains from copying. They make initial authorization, purpose limits, and collective bargaining more consequential.

A 2026 Harvard Business School working paper, still identified as preliminary, Labor as Capital: AI and the Ownership of Expertise, reports that when workers know their work data may train AI to perform similar work, they provide less existing data and demand a higher reservation wage for continued labor. The finding should be treated with the caution appropriate to a preliminary working paper and does not establish the response in every industry, but it illustrates the incentive problem: a rule intended to maximize company access can reduce the production and sharing of useful context.

Absolute individual ownership fails too. If every item of work context belongs wholly to the employee who touched it, companies will invest less in training and knowledge infrastructure. A key employee could hold a team's jointly built operational capacity hostage. The privacy of customers and colleagues could be mistaken for an asset the worker is entitled to sell. Companies could become unable to satisfy audit, regulatory, security, record-retention, and business-continuity obligations.

There is also a worker-to-worker externality. Allowing each worker to sell context independently does not necessarily empower labor. One person's sale can improve the company's capacity to automate or substitute for other workers, weakening their bargaining position. Workers may then underbid one another for a payment that does not reflect the collective loss of leverage. Team context is therefore often better governed through collective bargaining, a knowledge guild, a union, a data trust, or another pooled institution than through individual micropayments alone. The companion HBS paper Knowledge Guilds proposes collective administration of training uses and returns through such institutions. Collective bargaining does not erase individual contribution; it manages worker-to-worker effects that a bilateral sale cannot internalize.

The political-economic issue is not whether investment deserves a return. Companies supply capital, data infrastructure, coordination, market access, legal responsibility, and risk-bearing; those contributions warrant protection and reward. The issue is whether one employment payment should silently purchase an unlimited, perpetual, transferable right to reproduce a human capability, including after employment ends. At the other extreme, it is equally untenable to let an individual privatize the contributions of teammates, the employer's protected environment, or third-party data.

A sustainable regime must make continued knowledge production rational for both sides. It should protect the firm's factual and operational assets, preserve the worker's future productive agency, recognize collective production, and price uses that exceed ordinary employment, especially continued operation of a personalized model after departure.

What Belongs to the Individual—and What Belongs to the Company

The following matrix treats control as a bundle of rights rather than a single absolute title.

The decisive fact is not which computer was used to create an artifact. Four questions provide a better boundary test:

1. Does the capability still work after company names, people, numbers, internal thresholds, and systems are removed? If yes, it leans toward the individual as a general capability.
2. Can it reveal or reconstruct customer identities, trade secrets, internal thresholds, credentials, tool maps, or organizational relationships? If yes, it leans toward company or third-party stewardship, or it must be excluded from export altogether.
3. Was it produced jointly by personal judgment, team collaboration, and company material? If yes, licenses should be split, provenance should be preserved, and major uses and revenue should be governed jointly rather than assigned absolutely to one side.
4. After the worker leaves, does the company continue operating a model that substantially copies that worker's general judgment, style, or identity? If yes, the company must obtain independent consent and agree to an auditable remedy: deletion, depersonalization, a one-time buyout, or a continuing royalty.

These tests distinguish portability from disclosure. A portable skill core should encode a general method without carrying the company's factual world. A contribution attestation should let a former worker prove what they helped accomplish without exposing confidential evidence. Conversely, the company should be able to retain operational continuity, fixed deliverables, protected facts, and auditable records without claiming permanent ownership of every general capability that arose during the relationship.

The matrix is normative, not a statement that every current jurisdiction already applies these outcomes. Existing statutes, contracts, collective agreements, professional duties, and case-specific facts can change the legal result. Its purpose is to identify the institutional distinctions that become necessary if context is the productive capital through which a person can work with ASI.

A Context Bill of Rights for the Future

A future Context Bill of Rights should establish a workable constitutional settlement between personal agency, enterprise continuity, collective production, and third-party rights. It must operate at entry, during routine training, when a new purpose is proposed, when a model may replace people, and when a worker exits; a promise that applies only after a dispute is too late.

Individuals should receive at least six operational protections:

• Cognitive continuity: continued access to a personal core and to a professional-capability representation that does not disclose protected material.
• Semantic portability: export not merely of a PDF, but of structure, provenance, version history, schema, and a callable skill representation, subject to sanitization against secrecy and reconstruction risks.
• Background-asset registration: the ability to register a pre-existing knowledge base, skill, model, and public method when entering employment so a broad assignment does not silently absorb them.
• Contribution attestation: verifiable proof of projects, roles, judgments, and results that can travel without copying the company's raw evidence.
• Digital-double control: absent independent, revocable consent, a company should not keep acting externally under a worker's name, voice, style, or substantive judgment persona. On exit the options should include deletion, depersonalization, a buyout, or a continuing paid license.
• Audit, correction, and appeal: the ability to know which context is captured, trained, inferred, and used for which decisions, and to challenge errors, new purposes, or automated decisions.

Companies also need defined rights: control of original business records, trade secrets, credentials, proprietary implementations, and company bindings; rights to modify and continue operating express work products; retention and audit powers needed for security, regulation, litigation holds, customer contracts, and continuity; and an explicit, purpose-limited, non-exclusive license to call a personal skill core during employment. The right to train AI on a worker's context should require authorization separate from the ordinary employment agreement rather than being buried in an unlimited assignment clause.

For jointly produced context, major training, new purposes, personnel substitution, and revenue allocation should involve a worker council, union, knowledge guild, or context data trust. A personalized model used after employment should be deleted, depersonalized, bought out, or subject to a continuing royalty. Team context should generate a collective revenue pool rather than being deemed valueless by default or settled token by token.

The portable compiler described below is a normative institutional proposal, not a capability or legal entitlement that current law generally guarantees. Technically, it would implement the boundary as follows:

personal_skill_core
        -> controlled projection
company_context + company_binding
        -> sanitization, audit, approval
portable_skill_core + portable_attestation

The principle is that an individual's capability may enter the company boundary and operate there, while the company's raw context does not automatically flow back into the individual's knowledge base. At exit, a controlled compiler removes names, facts, internal numbers, customer references, thresholds, schemas, credentials, and tool bindings that belong inside the company domain. It then tests the remaining representation for reconstruction risk and produces a portable general skill plus an attestation of contribution. Every derivative should retain source labels, provenance, license, purpose, sensitivity, and re-export rules. The process needs logs, review, approval, and a channel for challenge; "sanitized" cannot be a label applied by either side without evidence.

If context truly contributes 99% of productive value, compensation should not be limited to a one-time wage. A more complete model can contain five parts:

1. A wage for current labor: payment for present time, attention, execution, and responsibility.
2. A background-context license fee: a purpose- and duration-specific license when the company calls context that the worker brought into the job or continued to develop independently.
3. A digital-double buyout or royalty: separate consent, pricing, and audit when a highly personalized judgment model remains in use after the worker leaves.
4. A pooled context dividend: a collective dividend, profit share, or equity interest for inseparable context produced by a team rather than a default assumption that its contribution is worth zero.
5. A return on company investment: a legitimate return for capital, data, organizational coordination, brand, risk, and legal responsibility rather than treating either side's contribution as a free raw material.

Tracking the marginal contribution of every token is usually infeasible and may create a false precision that benefits the party controlling the meter. A more realistic approach is to define categories, purposes, duration, and revenue-sharing rules in advance; place team-created value into a collective pool; and use independent audits to verify which models, data classes, and capabilities are actually being used.

This proposal distinguishes current law from new normative rights. The cognitive continuity right, a special digital-double right, and a context dividend are institutional proposals, not rights that are already generally enacted. Their justification rests on the possibility that context will become both an extension of the person and the principal productive asset through which labor participates in an AI economy.

The settlement can be stated in one sentence:

A person should be able to carry away a functionally equivalent self, but not the company’s factual world. A company may retain its secrets and work product, but it should not turn a worker’s future professional identity into a permanent, uncompensated corporate asset.

Principal sources

• Andy Clark and David Chalmers, The Extended Mind
• Margaret Jane Radin, Property and Personhood
• Barth, Datta, Mitchell, and Nissenbaum, Privacy and Contextual Integrity
• Stanford Encyclopedia of Philosophy, Republicanism
• 17 U.S.C. Section 201, Ownership of Copyright
• 17 U.S.C. Section 102, Subject Matter of Copyright
• 18 U.S.C. Section 1839, Definitions for Trade-Secret Law
• Morlife, Inc. v. Perry
• EU Trade Secrets Directive 2016/943
• California Labor Code Section 2870
• European Data Protection Board, Guidelines on the Right to Data Portability
• International Labour Organization, Navigating Workers' Data Rights in the Digital Age
• EU Platform Work Directive 2024/2831
• Gary Becker, Investment in Human Capital
• Grossman and Hart, The Costs and Benefits of Ownership
• Michael Polanyi, The Tacit Dimension
• Harvard Business School Working Paper 26-063, Labor as Capital: AI and the Ownership of Expertise
• Harvard Business School Working Paper 26-064, Knowledge Guilds

Research boundaries

• The ASI timetable, the relative capacities of humans and AI, and the 99% figure are stipulated assumptions for a thought experiment. They are not facts established by this essay.
• The legal discussion is structural analysis, not legal advice for any particular contract, jurisdiction, person, or dispute.
• The terms "cognitive continuity right," "digital-double right," "portable compiler," and "context dividend" name normative institutional proposals. They are not legal rights that currently exist in general form.
• The questions that most need empirical testing are whether context's marginal value can be measured, which forms of portability avoid leakage, the actual replacement effect of personalized models, and the transaction costs of collective governance.