AI and the Law

The Jurisprudential Problem of Artificial Computation

Law organizes public authority through norms that determine what persons may claim, what institutions may decide, what conduct is forbidden or required, and what remedies follow from a wrong. Artificial intelligence changes the conditions under which these norms are formulated, applied, interpreted, and enforced. An AI system may recommend the classification of a welfare recipient, determine what evidence an investigator sees, rank a candidate for employment, draft a contract, produce the reasons that appear in an administrative decision, or enable a service to operate across borders without a conventional institutional center. Each operation takes place within legal relations that already exist. The novelty arises when inferential and generative capacities become intermediate structures through which people encounter law.

The philosophical problem is therefore the relation between legal normativity and computational mediation. Legal institutions ordinarily claim that their decisions are authorized by valid sources, justified by recognizable reasons, attributable to responsible agents, and open to correction through established procedures. Computational systems introduce a distinctive difficulty: their operational determinants can be distributed among developers, datasets, model weights, interfaces, organizational incentives, procurement arrangements, and users. A legal conclusion may be presented as a single administrative act even though the conditions producing it are partly outside the deciding institution's direct knowledge. The problem is structural, because merely identifying the person who clicked a final confirmation button does not disclose where relevant choices were made.

This monograph investigates the conceptual grammar of law under those conditions. Its principal thesis is that the authority of AI-mediated legal action depends on the preservation of legally intelligible relations among source, decision, justification, attribution, and remedy. These relations require interpretation, institutional design, and public accountability at different stages. The proposal is an original jurisprudential synthesis, not a description of a single enacted legal regime. It distinguishes the validity of the applicable law from the legality of its computational implementation, the accuracy of a prediction from the sufficiency of a legal reason, the availability of information from effective contestability, and the philosophical category of an artificial bearer of reason from positive-law personhood. The historical foundation comes from jurisprudence; the contemporary evidence comes from legal texts, reported judgments, and scholarship on technology and procedure (Hart, 1994; Raz, 1979; Citron, 2008; Hildebrandt, 2020).

Definition, Scope, and Levels of Inquiry

AI and the Law is the systematic philosophical investigation of how artificial intelligence interacts with the nature, content, authority, administration, enforcement, and development of legal norms and legal institutions. It includes the application of established legal rules to AI activities, the incorporation of AI systems into legally consequential decisions, and the conceptual question of whether established categories adequately describe emerging artificial actors and forms of provenance. These dimensions are related but analytically distinct. The first asks what law governs AI; the second asks what happens when law operates through AI; the third asks how AI challenges or clarifies the legal vocabulary itself.

An AI system is understood here as a machine-based inferential system capable, under varying degrees of autonomy, of generating predictions, recommendations, decisions, or content that influence an environment. This description accommodates statistical classifiers, large language models, generative systems, and agentic arrangements. It does not assume that all such systems reason in the same way or possess a common moral standing. The OECD's revised Recommendation is a useful cross-jurisdictional terminological reference; its 2024 update also reflects growing attention to general-purpose systems, information integrity, intellectual property, and safety (OECD, 2024). A technical classification, however, cannot itself determine what obligations legal institutions should recognize.

The fundamental unit of inquiry is an AI-mediated legal practice: a configuration of human actors, computational processes, organizational procedures, and authoritative norms whose interaction has legal consequences. Jurisprudence examines what makes the configuration legally valid and justified. Public law examines its authority and constraints. Private law allocates claims, duties, and losses. Procedural law defines proof and remedies. Constitutional and human-rights law set boundaries against arbitrary interference. Each branch contributes a different part of the explanation.

Some neighboring dossiers are reserved for a more specific object. Algorithmic Adjudication investigates judicial decision procedures. Machine-Generated Legal Decisions examines the status and provenance of generated rulings and documents. Legal Accountability for AI Systems concentrates on liability. Legal Personhood examines the legally conferred status of a right-and-duty bearer. This work integrates those questions only insofar as their relationship is necessary to understand AI's general transformation of legal concepts. Its object remains the jurisprudential structure connecting computation and law.

Ancient Foundations: Norm, Judgment, and Practical Reason

Aristotle's analysis of justice supplies an enduring distinction between general rules and the judgment required for their application. In the Nicomachean Ethics, Book V, distributive and corrective justice concern different patterns of allocation and rectification. His discussion of equity, epieikeia, at V.10 (1137a31–1138a3), addresses the generality of law: a universally formulated rule can encounter a case whose relevant circumstances exceed the rule's literal scope. Equity corrects the application of a universal formulation where its generality makes it inadequate. The point is foundational for AI and law because computational classification often succeeds by imposing stable categories at scale, while adjudication sometimes requires identifying why the particular situation makes the general category incomplete.

This is a philosophical analogy, not a claim that Aristotle anticipated machine learning. The historical argument concerns the necessary relation between rule and application. A model that correctly identifies most cases according to an administratively defined category can still misclassify the legal importance of an exceptional case. If the legal institution has no procedure for recognizing the exceptional facts, high aggregate accuracy does not remedy the jurisprudential failure. Aristotle's analysis supports a demand for practical judgment located within a responsible legal process rather than a general rejection of rules (Aristotle, Nicomachean Ethics, V.10).

Roman juristic thought developed another important foundation: legal relations are structured through recognized persons, things, and actions, while procedural forms define how claims are made effective. The Institutes of Gaius, particularly Book I on persons and Book IV on actions, records a legal vocabulary that cannot be reduced to physical or psychological attributes. The legal capacity of a participant depends on the normative structure of the applicable order. This history illuminates contemporary discussions of artificial entities: asking whether an artificial system is a legal person is a question about the legal consequences of recognition, the distribution of claims and responsibilities, and the institutional rules that authorize recognition. It is not answered by naming the technical sophistication of the system alone.

Early Modern Jurisprudence: Sovereignty, Rules, and Authority

Hobbes describes civil law in Leviathan, Chapter 26, through the authority of the sovereign and the public establishment of rules. Law serves as an authoritative standard for those subject to common political power. His argument illuminates the problem of attribution when computational systems assist a state: a technical supplier may provide the mechanism, but a decision that exercises public power must remain traceable to an institution possessing lawful authority. A state cannot fully explain an exercise of coercion by identifying the vendor from which it purchased its software.

Bentham and Austin sharpened the institutional identification of law in their command-oriented approaches, while subsequent jurisprudence exposed the inadequacy of reducing all legal phenomena to commands backed by sanctions. Contracts, wills, corporate formation, procedural rights, and delegations of administrative power are among the law's power-conferring structures. Computer systems now commonly implement such legal powers, for example by recording a contractual assent or executing a condition of an electronic transaction. A theory focused only on prohibitions and penalties would overlook the legal significance of who can validly alter another's normative position.

Kant contributes a different basis in the Doctrine of Right, part of The Metaphysics of Morals (1797). External freedom must be capable of coexistence under a universal law, while coercive authority requires a public juridical condition. The significance of this framework is that legal technology operates among persons whose freedom must be reciprocally secured. A risk classification that optimizes collective efficiency at the expense of an individual's standing cannot be justified merely by its utility. Kant's position offers a normative criterion for examining whether public computational practices treat affected individuals as parties to a common legal order.

These historical accounts place three requirements in view: an ascertainable source of authority, an account of the kinds of normative powers that legal rules create, and a justification for publicly binding persons. Contemporary AI regulation must address all three. A decision-support system can be technically reliable yet exceed delegated powers, alter the practical effect of a right, or compromise equal standing.

Legal Positivism and the Sources of Computationally Administered Law

H. L. A. Hart's distinction between primary and secondary rules in The Concept of Law remains productive for AI-mediated institutions. Primary rules guide conduct; secondary rules establish procedures for recognizing, changing, and adjudicating the rules of an order. Hart's rule of recognition is a social practice among officials that provides criteria for identifying legal validity. Computational processes may assist in searching, organizing, or applying recognized sources, but statistical association with prior decisions does not itself become a criterion of legal validity. A model trained on judicial opinions can reproduce prevailing patterns without possessing the institutional authority that makes a statute or precedent a source of law (Hart, 1994, chapters V–VII).

This difference becomes practical when an AI system produces a plausible answer unsupported by the relevant legislature or court. The answer may be persuasive to a user, but its legal force depends on an authorized source and a valid interpretive process. A legal institution that silently substitutes generated propositions for enacted norms risks confusing informational confidence with official recognition. The problem cannot be solved by more accurate text generation alone: a perfect paraphrase of a repealed rule remains a paraphrase of a rule that lacks present validity.

Positivism also helps separate a claim about what the law currently requires from an argument about what it should require. A statute may create a legal obligation to audit certain AI applications even if philosophers disagree about whether the statute goes far enough. Conversely, an excellent ethical code may have no binding legal force unless the applicable legal order incorporates it. The contrast permits serious evaluation of soft-law frameworks such as the OECD Recommendation and UNESCO's ethics Recommendation without misdescribing them as universally binding legislation.

The principal objection is that interpretation of apparently source-based requirements often invokes values, purpose, proportionality, and rights. Modern positivism can accommodate this problem in competing ways; the dispute between exclusive and inclusive versions concerns the conditions under which moral standards figure in legal criteria. AI makes the dispute visible because machines increasingly transform broad legal language into operational thresholds. The law's validity, its content, and the operational implementation of that content remain distinct questions.

Fuller and the Conditions of Legality

Lon L. Fuller argues in The Morality of Law that a legal order depends on formal conditions permitting rules to guide conduct. His eight principles concern generality, publicity, prospective operation, intelligibility, consistency, feasibility, relative stability, and congruence between declared rules and official action. The principles describe what an institution must do if it claims to govern through law rather than unstructured coercion (Fuller, 1965). Artificial systems bring each condition under pressure because a policy may be announced publicly while the operative decision procedure remains inaccessible, unstable, or different from the official rule.

Consider a public agency that promulgates eligibility criteria for assistance but relies on a machine-learning score to classify applications. If the threshold was never authorized, the actual decision practice diverges from the published law. If the model changes through updates without recorded oversight, the criteria governing comparable applicants may become unstable. If people cannot learn the operative conditions under which their rights are decided, publication of a general statute may be insufficient to secure meaningful guidance. The gap between law in books and law in computational action can become a problem of legality itself.

Fuller's account does not imply that every model parameter must be publicly disclosed to every individual. Some systems require protection against misuse and may involve legitimate confidential information. The stronger inference is that persons and reviewing institutions need access to the legally material rules, assumptions, performance conditions, and reasons. The exact form of access can vary with the nature of the case. An unexplained proprietary restriction cannot settle what a constitution, statute, or fair procedure requires.

A possible objection holds that Fuller's principles concern governing rules, while probabilistic models simply provide evidence or advice. The distinction is valid where the model is genuinely evidentiary and independently assessed. But when an output determines the operative threshold for approval, prioritization, investigation, or sanction, the model participates in practical rule administration. The institutional task is to disclose the point at which the tool becomes functionally authoritative.

Dworkin, Interpretation, and Justification

Ronald Dworkin's legal interpretivism treats legal reasoning as an interpretive practice in which institutional history and political morality jointly matter. Law as Empire (1986) develops the idea of law as integrity: adjudicators should understand the legal record through principles that provide a coherent and justified account of the community's legal practice. This picture challenges the assumption that judicial decisions can be reconstructed as direct deductions from isolated rule statements. Precedent, purpose, equality, and principle shape the interpretation of legal materials.

The implications for large language models used in legal research are significant. A generated answer can retrieve authorities and synthesize doctrinal patterns, but it can also omit the interpretive premise that connects an authority to the case. Two superficially similar holdings may depend on different constitutional purposes or institutional contexts. Their semantic proximity does not determine whether they support the same legal conclusion. A model that produces a correct citation without understanding or exposing the principle connecting it to the proposition supplies an incomplete legal reason.

Dworkin's account also makes visible disagreement about the meaning of a legal source. Where reasonable legal professionals contest a principle, an AI tool that collapses the disagreement into one confident response may create a false appearance of settled law. The requirement is therefore to distinguish retrieval, doctrinal synthesis, and normative interpretation. A human legal officer remains answerable for an adopted position, but that answerability has substance only if the officer can evaluate competing interpretations.

A critical issue arises: Dworkin's ideal of coherence itself depends on contestable value judgments. Computational assistance may make the available cases more legible and expose inconsistencies, yet there can be no neutral optimization criterion that determines every weight among historical fit, equal treatment, constitutional purpose, and institutional stability. Legal reasoning retains an irreducible element of public argument.

Raz: Authority, Guidance, and the Limits of Delegation

Joseph Raz's work on legal authority examines the conditions under which an institution can legitimately claim to guide action. The Authority of Law (1979) distinguishes legal authority's claims from the question of whether citizens always have a moral duty to comply. His account of authority raises a focused question for computational governance: can an automated recommendation serve as a practical intermediary through which an institution provides reasons that those subject to it can understand and assess?

A model's performance is relevant, but performance alone does not create legal authority. A reliable medical system may justify a recommendation because of its evidential credentials, while the authority to deny treatment under public insurance law depends on a legal mandate, relevant rights, and authorized procedures. Those with lawful decision-making power may use technical expertise as a premise. They cannot thereby dissolve the distinction between evidence for a judgment and power to bind another person.

The institutional argument can be reconstructed in four steps. First, law identifies which actors or offices have the authority to alter legal positions. Second, AI produces informational outputs by processes configured for a particular use. Third, such outputs become legally consequential only through their incorporation into authorized practices, contracts, or official acts. Fourth, the validity and legitimacy of that incorporation depend on the underlying legal standards, including restrictions on delegation and opportunities for challenge. The argument makes accountability a relation established by law, rather than a property inferred from the apparent intelligence of the tool.

The counterargument emphasizes that modern bureaucracies routinely rely on experts whose reasoning no official fully reproduces. That is true, and it means interpretability cannot be demanded without regard to the type of delegated expertise. Yet an institution must still determine when expertise has exceeded its mandate, what facts were relied upon, and who can correct mistakes. AI does not invent this demand; it intensifies an existing problem of legal authority.

Hohfeld: Rights, Duties, Powers, and Immunities

Wesley Newcomb Hohfeld's Fundamental Legal Conceptions as Applied in Judicial Reasoning (1913; 1917) demonstrates that legal language becomes confused when different normative relations are grouped under the undifferentiated word right. Claim-rights correlate with duties; liberties with others' no-rights; powers with liabilities to normative change; immunities with the disabilities of other actors to effect that change. His analysis is especially useful for AI law, where public discussion often attributes a supposed right, entitlement, ownership, or freedom to an AI system without specifying its correlative legal relation (Hohfeld, 1917).

When a statute grants a data subject a claim to receive information about automated processing, a corresponding duty must be identifiable. A developer may have a duty to document a system; a deployer may owe an explanation to an affected person; a supervisory authority may possess the power to order changes. These are not interchangeable roles. Likewise, a software agent may possess a technical capability to transmit an order without possessing an independent legal power to form a contract in its own name. Legal effect may arise through rules of agency and attribution to the person or corporation on whose behalf it operates.

Hohfeld's framework also disciplines arguments about artificial personhood. The creation of an entity with limited contractual capacity, property-holding capability, or standing to sue would entail specific legal powers, duties, and immunities. It would not automatically confer moral status, political participation, or every incident of citizenship. Conversely, the lack of general personhood does not imply that an artificial system's outputs are outside law; their producers, operators, users, and affected parties remain situated in numerous legal relations.

The analytical advantage is precision. Proposals to recognize a new category of artificial entity should specify the legal incidents being created, the persons protected by them, and the institutions entitled to enforce them. An abstract declaration of status without such specification risks concealing rather than resolving normative problems.

Legal Realism, Institutional Practice, and Computational Prediction

American legal realists emphasized the significance of judicial behavior, administrative routines, and the social consequences of legal rules. Their work invites a distinction between the authoritative formulation of a norm and the observable pattern of its application. Statistical models often learn the latter. Predicting whether a court is likely to uphold a claim is a different activity from interpreting the law governing that claim. The distinction matters because a system trained on historical outcomes may reproduce inequalities and institutional habits that legal reform seeks to overcome.

Suppose a prediction model forecasts that defendants from a certain neighborhood are more likely to receive pretrial detention. Even if historically accurate, that prediction does not establish that neighborhood is a legitimate legal reason for detention. A realist account helps investigators locate the practical site of decision power, while normative legal analysis determines whether that practice respects the applicable standard. The two tasks are complementary; confusing them can convert an empirical record of institutional conduct into a pseudo-legal command.

Realist insights support empirical auditing of who uses a model, when officials override it, what decisions its output actually changes, and whether formal review procedures alter outcomes. They also reveal how interfaces and workload pressures can reshape an institution without any formal amendment of the law. A recommendation that administrators nearly always accept may acquire de facto authority despite nominally advisory status.

The critical danger is that prediction can normalize what ought to be contested. The philosophical solution is to preserve two distinct questions in institutional records: what outcome does the system expect under current conditions, and what outcome do legally justified principles require? Legal reform often begins precisely where the answers diverge.

Critical, Feminist, and Sociolegal Perspectives

Critical legal studies and feminist jurisprudence investigate the ways apparently neutral rules may reproduce structured inequalities. AI classification adds a new layer of potential mediation: decisions can be made through historical datasets carrying unequal access to resources, discriminatory enforcement patterns, and omissions from public records. A variable can be legally permissible in abstract form yet function as a proxy for a protected characteristic within a particular deployment. Neither removal of a protected column nor aggregate model calibration conclusively establishes nondiscrimination.

The epistemic problem concerns who defines the target. If a model is trained to predict prior enforcement actions, it may learn an institution's past decisions rather than independently existing misconduct. If its objective is to reduce administrative cost, efficiency may be achieved by limiting the opportunity of difficult claimants to explain atypical circumstances. These choices are legal and political decisions expressed through technical systems, not value-free discoveries.

Feminist and critical perspectives direct attention to the distribution of procedural burdens. A claimant may technically have an appeal yet lack the knowledge, time, language access, or legal assistance necessary to use it. Statistical fairness metrics evaluate patterns among groups under formal definitions; lived procedural justice requires the capacity of actual persons to challenge consequential errors. Institutional evaluation must therefore include accessible reasons, representation, correction procedures, and the unequal costs of contestation.

The strongest objection to this approach is that human decision-making also exhibits prejudice, inconsistency, and opacity. Comparative analysis should acknowledge that automated systems can sometimes improve consistency or reveal hidden patterns. The relevant question is whether a particular arrangement produces an accountable improvement over realistic alternatives, measured across lawful outcomes, affected groups, and the ability to seek correction. No abstract preference for human or machine decision-makers settles the inquiry.

From Legal Informatics to General-Purpose AI

The history of AI and law includes expert systems, formal representations of statutes, automated retrieval, case-based reasoning, prediction, machine learning, and generative language models. Early rule-based approaches attempted to encode conditions explicitly. They often exposed the difficulty of translating open-textured legal categories into precise computational terms. Later statistical systems learned regularities from examples rather than relying on fully articulated rules. Large language models transformed the field again by making plausible legal prose and interactive research interfaces broadly accessible.

These developments change the relationship between computational output and its grounds. In a hand-built rules engine, a decision may be linked to an explicit logical condition, although the choice of rule and its representation still requires scrutiny. In a statistical classifier, an output may depend on numerous learned correlations whose normative significance is uncertain. In a generative model, a coherent explanation can be produced independently of a reliable chain from authoritative source to proposition. The fact that an answer is expressed as a legal argument provides no guarantee that it was generated through legally appropriate reasoning.

The history also includes a persistent aspiration to make legal knowledge accessible. Improved translation, document search, and argument organization can reduce barriers faced by people without specialist representation. These gains must be evaluated alongside the risk that generated errors may be especially difficult to detect for those least able to hire a lawyer. Access to language is a contribution to legal access only where legal accuracy and correction remain attainable.

Mireille Hildebrandt's Law for Computer Scientists and Other Folk (2020) frames law as a practice of institutional and interpretive significance rather than a set of computational instructions. Her treatment of legal personhood, rule of law, and protection by design illustrates why constructing systems that operate within legal orders requires understanding legal forms and remedies, not simply translating statutes into code.

The Normative Translation Problem

A legal standard is a directive whose meaning is established within an institutionally maintained practice. A computational rule is an executable specification mapping inputs, states, or learned representations to outputs. Their relation requires interpretation and design. A legal requirement such as reasonable care, proportionality, best interests, or appropriate explanation leaves room for judgment about purpose, context, conflicting rights, and relevant evidence. Any implementation operationalizes a particular interpretation of these conditions.

Translation can occur at several points. A legislature selects an objective; an agency formulates policy; procurement staff write requirements; developers choose target labels and data; auditors select metrics; officials determine how strongly to rely on results; and reviewing bodies define acceptable proof of compliance. Legal choices can be made at every stage even when the system is described as purely technical. The practical exercise of law may therefore be dispersed before the formally accountable official reaches a decision.

This yields a useful distinction. Source validity concerns whether a norm has been established according to the legal order. Semantic determination concerns what that norm means. Operational translation concerns how a system implements its requirements. Case application concerns whether the result is justified on the facts. Institutional review concerns how mistakes or unlawful practices are corrected. Success at one level cannot substitute for proof at another. A valid statute may be implemented by an unlawful classifier; a technically correct classifier may be deployed beyond an agency's authority.

A legal design process should preserve documentation of the translation steps, including rejected alternatives and the legal authority for contested choices. Such documentation need not expose protected security details publicly, but it must be sufficient for competent oversight. This is the difference between treating implementation as private engineering and recognizing it as an institutional act capable of changing the practical meaning of rights.

Rule of Law, Publicity, and Predictability

The rule of law requires that public power be constrained by law that is capable of guiding conduct and subject to appropriate forms of review. In AI-mediated administration, formal publication of a statute can coexist with opaque operational decision rules. A person may know the stated eligibility requirements but remain unable to discover why a system assigned a risk score or denied an application. If the score has decisive effect, practical predictability and equality before the law can suffer.

Predictability has several senses. People need adequate notice of the legal rule. Officials need stable criteria for its application. Courts need a reconstructible record against which they can test legality. Developers need specification sufficient to implement the rule as authorized. A system's ability to predict human behavior or likely administrative outcomes is not equivalent to the public's ability to predict how legal power will be exercised. The philosophical mistake is to equate statistical forecasting with legal certainty.

Explanations serve different audiences. The affected individual needs to understand the reasons material to their own case. The regulator requires information on systemic performance and compliance. A court requires evidence sufficient to decide contested issues. The public needs to know which kinds of consequential decisions are automated and under what authority. These audiences may appropriately receive different technical detail. A single disclosure statement cannot replace all the institutions of legality.

Fuller's congruence requirement and Raz's analysis of law's guidance function converge on a practical principle: governmental use of AI must remain consistent with the rules under which the government claims authority. Algorithmic optimization can improve administrative coordination, but legality demands an intelligible connection between declared norms and operative practices.

Due Process as an Institutional Relation

Due process protects a participant's standing to receive notice, present a case, challenge adverse material, and obtain an impartial decision according to applicable rules. Its content differs among legal orders and procedural settings. The central philosophical idea is that a person affected by authoritative action is entitled to more than a statistical prediction of their likely outcome. They are entitled to the legally established modes of participation and review appropriate to the decision.

Danielle Keats Citron's Technological Due Process (2008) identified the danger that automation can collapse rulemaking into adjudication while securing the safeguards of neither. Administrative code may quietly encode policy choices, and individual hearings can lose their significance when decision-makers regard computational outputs as presumptively infallible. The difficulty is institutional rather than merely explanatory: who can change the rule, who can see that it changed, and where can the affected person challenge it?

Due process analysis should investigate notice of AI use, disclosure of materially adverse information, opportunity to correct inaccurate data, access to expert assistance where required, independent reconsideration, preservation of a decision record, and authority to order a remedy. A procedure is effective when it can change a decision or the rule producing similar decisions, rather than only acknowledge dissatisfaction. The availability of a human reviewer is therefore a beginning. The reviewer must possess legal authority, relevant information, sufficient time, and an institutional incentive to disagree when disagreement is justified.

The counterargument points to mass administration: requiring highly individualized procedures in every low-stakes transaction may delay benefits and divert scarce resources. Legal systems already calibrate procedural safeguards according to context. The jurisprudential requirement is to identify the stakes and reviewability of the decision, then allocate procedures capable of protecting the rights at risk. Administrative efficiency cannot silently determine the rights themselves.

Evidence, Explanation, and the Burden of Proof

Legal evidence becomes relevant through rules about material facts, reliability, admissibility, disclosure, burdens, and standards of proof. An AI output may serve as evidence, as a tool for locating evidence, or as an element of a decision. These roles demand separate evaluation. A prediction that a defendant will reoffend is not direct evidence that the defendant committed the charged offense. A generative summary of a document is not a substitute for the authentic record. A system's asserted confidence is not a legal standard of proof.

Explainability similarly has several layers. A technical explanation may identify input features or local model behavior. An institutional explanation identifies the purposes for which the system was adopted and the standards controlling its use. A legal reason identifies facts and norms sufficient to justify the decision. A litigant may need a chain linking all three, but no single feature-attribution diagram supplies an authoritative legal interpretation.

The burden of proof becomes especially difficult when a claimant can observe an adverse result but relevant model evidence is possessed by a private vendor or public agency. Trade secrets, national security, and privacy can legitimately affect disclosure, yet they do not automatically justify depriving a claimant of effective remedies. Protective orders, confidential expert examination, independent audits, retention requirements, and carefully specified presumptions may help reconcile conflicting interests. The choice of mechanism must be justified within the procedural law at issue.

A philosophically adequate standard asks whether the evidence permits a reviewing body to resolve the contested legal proposition. It does not demand complete mathematical transparency for its own sake. Conversely, a technically impressive report cannot replace proof of the facts and legal criteria actually in dispute.

A Jurisprudential Reading of State v. Loomis

In State v. Loomis, 2016 WI 68, the Wisconsin Supreme Court considered the use of a COMPAS risk assessment in sentencing. The court upheld consideration of the assessment subject to limitations and cautions; it did not authorize COMPAS to determine a sentence by itself. The case illustrates how risk prediction can enter an authoritative process while the deciding judge retains formal responsibility. It also exposes a recurring difficulty: proprietary methods can limit a defendant's ability to scrutinize the basis of a reported score (State v. Loomis, 2016).

The first philosophical question is the relevance of the predicted object. Reoffending risk is a probabilistic estimate of future behavior, while sentencing traditionally involves multiple legal considerations including culpability, proportionality, public safety, and rehabilitation. A statistical output cannot decide how these normative considerations should be weighted. The second question concerns evidentiary reliability: population-based correlations may be relevant to an identified purpose without accurately representing an individual's circumstances. The third concerns procedural equality: a defendant's opportunity to challenge a score may be limited by inaccessible assumptions and training data.

It would be inaccurate to present Loomis as a general constitutional prohibition on predictive systems, or as an unqualified endorsement of them. The holding is more specific, and its philosophical importance lies in the tension between using information and granting it legally decisive force. The case supports careful examination of the place of computational output within a larger legally constrained judgment. Whether later uses meet those constraints must be evaluated separately under their own record and applicable law.

Automated Administration and the SyRI Judgment

The Hague District Court's 2020 judgment concerning the Dutch System Risk Indication, SyRI, examined a governmental system intended to detect risks of social-security and related fraud. The court held that the legal framework for SyRI violated Article 8 of the European Convention on Human Rights, addressing the lack of adequate safeguards and the balance between public interests and private life (District Court of The Hague, 2020; Rechtspraak, 2020).

The legal significance extends beyond prediction error. A model may be effective at identifying suspicious patterns yet still be employed under powers insufficiently bounded by privacy and proportionality requirements. The court's intervention concerned the law authorizing and regulating the practice. The case therefore shows that legality may be tested at the level of systemic design and statutory authorization, not only by the outcome of a particular individual's risk score.

Philosophically, SyRI reveals the possibility of broad governmental surveillance through the accumulation and recombination of datasets. Administrative categories designed for different purposes can be linked into a new regime of suspicion. Individuals may become subjects of investigation through correlations that are difficult to discover or contest. A law permitting such practices must be evaluated for the quality of its safeguards, the proportionality of its intrusion, and the distribution of burdens among groups.

The strongest defense of risk-based administration emphasizes limited public resources and the responsibility to detect fraud fairly. That defense requires evidence that the system is effective, authorized, necessary, proportionate, and appropriately monitored. The desire to prevent fraud is a legitimate public purpose in many legal orders, but a legitimate objective does not settle the legality of all means used to pursue it.

Facial Recognition and the Bridges Case

R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058 concerned live automated facial recognition used by police in public places. The Court of Appeal held that the legal framework governing important discretionary choices lacked the necessary quality of law, and it found deficiencies involving data protection assessment and the public sector equality duty. The judgment illuminates how computational systems can make governmental discretion both more scalable and less visible.

A central issue was the operational selection of watchlists and deployment locations. The existence of a technical capacity to recognize faces did not supply adequate limits on whose images could be compared, where the system could be used, and under which public rules those choices occurred. The case makes clear that procedural boundaries matter before the system generates an individual match. Unconstrained discretion can be exercised in system design and deployment policy rather than only at the final point of intervention.

The philosophical structure is a distinction among lawful public purpose, legal authorization, operational discretion, and rights-based constraint. A technology may contribute to crime prevention while its deployment remains legally defective. Equally, legal limits can be improved without treating the technology as intrinsically unlawful in every context. The governing question is whether public power is subject to accessible and sufficiently determinate constraints.

This problem is especially acute where a system processes people who have not chosen to interact with it. In public spaces, ordinary movements can become biometric events. Law must define the scope of the state's authority and the means by which individuals and oversight institutions can assess that authority. Public disclosure of a general surveillance program is not always sufficient to make its discretionary operation lawful (Bridges, 2020).

Data Protection and Automated Individual Decisions

Personal data protection is a major legal interface between AI and individuals. The EU General Data Protection Regulation establishes rules concerning lawful processing, purpose limitation, accuracy, transparency, and rights of data subjects. Article 22 concerns decisions based solely on automated processing, including profiling, that produce legal or similarly significant effects, subject to specified exceptions and safeguards. Its scope is narrower than a general prohibition on all automated assistance. Articles 13–15 and related provisions establish further information and access rights (Regulation (EU) 2016/679).

The Court of Justice of the European Union addressed automated credit scoring in SCHUFA Holding, C-634/21, judgment of December 7, 2023. It held that the automated establishment of a probability value can constitute automated individual decision-making under Article 22 when a third party draws strongly on that value to decide whether to enter, perform, or end a contractual relationship. The importance lies in functional influence: presenting a score as merely informational may fail to describe its practical decisiveness.

The philosophical lesson is that legal attribution must examine the causal role of the model within an institutional decision. A nominal human intermediary does not eliminate the question whether the score functionally determines the result. At the same time, Article 22's application depends on statutory conditions and cannot be extended by analogy to every AI-assisted act. Jurisdictions beyond the EU operate under different rules.

Data minimization also raises a conflict with demands for explanatory records. A system should retain sufficient material to permit lawful review without collecting unlimited personal data in the name of auditability. Legal governance therefore requires purpose-specific choices about retention, access, anonymization, and evidentiary preservation. Privacy and accountability must be designed as jointly governed interests.

Equality, Classification, and the Limits of Fairness Metrics

Legal equality requires examining the relation between the treatment of individuals and the criteria by which a legal institution distinguishes among them. AI systems classify persons using features selected for prediction or optimization. Statistical fairness research identifies multiple possible metrics, but those metrics embody different definitions and sometimes conflict. A classifier can have equalized performance by one criterion while imposing unequal errors or burdens under another. A legal equality analysis must begin with the governing rights, protected grounds, justifiable distinctions, and remedy.

Consider an automated hiring screen. It might equalize selection rates across defined groups while obscuring how qualification proxies correlate with disability or access to education. Alternatively, a model might improve individual prediction yet reproduce historical exclusion from particular occupations. Legal analysis must identify the legitimate purpose of the criterion, the relevant comparator, the distribution of mistakes, and the evidence needed to justify differential treatment. Mathematics assists but does not create the applicable legal standard.

This issue also involves procedural justice. A person may be wrongly classified because records are incomplete, names are inconsistently transliterated, or historical data reflect biased administration. A remedy needs an accessible means of correcting the record and challenging the legal use of the classification. Aggregate calibration cannot vindicate a person whose legally decisive facts were omitted.

Equality therefore operates at several levels: the validity of the classification purpose, the representativeness and provenance of data, the effects of model deployment, and the terms of review. Treating one global numerical score as a complete fairness verdict conceals the distinct legal questions the system raises.

Tort, Product Liability, and Distributed Responsibility

AI-associated harm may arise through defective design, inadequate warnings, negligent deployment, improper supervision, misuse, or failures to correct known risks. The relevant liability framework differs by jurisdiction and activity. Contract, negligence, strict product liability, professional duties, and statutory remedies can overlap. A philosophical investigation must distinguish responsibility for creating a risk, responsibility for operating a system, causal contribution to an injury, and the legal allocation of damages.

Distributed production complicates the proof of fault. A model provider may control training and updates; a fine-tuner may modify behavior; an application developer may design the interface; a deployer may set thresholds; and a user may act on the output. The existence of many participants can make the causal explanation more complex without establishing that no one is responsible. Law may place duties on different actors according to their control, foreseeability, contractual undertakings, and ability to prevent harm.

A common argument calls for strict liability whenever AI behavior becomes unpredictable. That proposal may strengthen compensation but can also discourage beneficial innovation or place costs on actors unable to manage the risk. Conversely, a universal fault standard may leave victims without meaningful redress when evidence of negligent design lies beyond their reach. A defensible regime must allocate risk in relation to control, public danger, available insurance, and the compensatory aims of the applicable legal order.

The proposed EU AI Liability Directive, COM(2022) 496, was withdrawn in October 2025; it must not be described as an enacted liability regime. The withdrawal does not eliminate existing tort, product, and other applicable obligations. It demonstrates why current-law analysis must separate legislative proposals, adopted instruments, commencement dates, and jurisdiction-specific remedies (European Parliament, 2025).

Contracts, Electronic Agents, and Attribution

Electronic contracting predates contemporary generative AI. The legal problems of offer, acceptance, agency, mistake, and authority already arise when software transmits or receives legally significant messages. Agentic AI adds the possibility of open-ended negotiation, tool use, procurement, and action under broad instructions. The question becomes how an external legal order attributes acts performed by software and whether that attribution respects the actual terms of authorization.

One possible model is ordinary agency: the software acts as an instrument for a principal, whose legal relations are affected within authorized bounds. Another involves direct statutory recognition of electronic agents as means of forming enforceable agreements, without granting those systems independent legal personhood. In either case, technical capacity to make an offer or transfer an asset is distinct from legal capacity to bind a specified party. Attribution depends on authority, applicable statutes, surrounding conduct, and contract terms.

Generative agents intensify the problem of exceeding instructions. If an AI assistant orders goods beyond its approved budget, the result may turn on apparent authority, contractual allocation of risk, notice to the counterparty, and consumer protections. A legal system must decide who bears the loss and what practices make the risk reasonably controllable. It cannot simply treat the machine's linguistic declaration as conclusive evidence that a principal consented.

At a conceptual level, contracts require coordination of normative powers across parties. AI can facilitate that coordination, but its role should be defined through verifiable authority boundaries, recordkeeping, revocation, error correction, and dispute resolution. These are legal design requirements regardless of whether the internal system is probabilistic or rule-based.

Criminal Law, Culpability, and AI Assistance

Criminal liability commonly depends on acts, prohibited consequences or circumstances, and required mental elements, subject to jurisdiction-specific doctrines and exceptions. AI can figure in the factual background of an offense, as a tool of a human offender, as an automated security apparatus, or as a source of evidence. These roles raise questions about intention, recklessness, negligence, causation, and proof, without establishing that every harmful computational output constitutes a criminal act by the system itself.

An operator might intentionally direct an AI service to commit fraud. A developer might knowingly facilitate a criminal scheme. A professional might negligently rely on erroneous automated advice in a setting where criminal negligence is recognized. Each scenario requires independent satisfaction of the law's elements. The mere fact that a model produced harmful content cannot supply an unexpressed mental element in the applicable statute. The analysis must identify whose act and state of mind the law treats as relevant.

The strongest philosophical reason for caution in attributing criminal guilt to a model is the relationship between culpability, institutional recognition, and penal justification. Punishment is not just a method of altering system behavior; it is an authorized imposition of a sanction under a legal standard. Where there is no legal category capable of receiving criminal liability, regulators may still require recalls, redesign, suspension, or sanctions directed at legally accountable persons or organizations.

Potential future recognition of artificial entities would not eliminate the need to justify a penalty's purpose and effect. A purely symbolic criminal conviction of a system that cannot bear or respond to the normative meaning of the sanction could obscure the responsibility of the people and institutions controlling its deployment.

Intellectual Property, Creation, and Legal Authorship

Generative AI raises questions about the origin of expressive works, legal ownership, infringement, training data, and the institutional reasons for granting exclusive rights. Copyright and patent systems do not share identical categories. Authorship in copyright, inventorship in patent law, moral attribution, philosophical originality, and publicly traceable provenance are conceptually different. Any claim that one of these has been established must specify the relevant legal order and legal test.

In Thaler v Comptroller-General of Patents [2023] UKSC 49, the UK Supreme Court held that DABUS did not qualify as an inventor under the relevant provisions of the Patents Act 1977. The decision interprets a statutory structure; it is not a metaphysical proof that artificial systems can never generate novel technical content. The United States Copyright Office's 2025 report on copyrightability similarly distinguishes human-authored expressive contributions from outputs attributed solely to generative systems in its interpretation of existing US copyright doctrine.

These materials show how legal categories can identify holders of recognized rights differently from the philosophical description of production. A work may have an artificial provenance and a discernible cultural authorial trajectory even when a jurisdiction's copyright rules confer protection only on particular human creative contributions. Conversely, disclosure of AI use does not determine whether another author's protected expression was unlawfully copied. Those questions must be investigated under copyright's own standards.

A sustainable jurisprudence of artificial creation should document provenance, roles, creative control, licenses, and permitted uses while keeping normative debates over future legislation separate from statements of existing law. The subject remains open to legal reform and cross-jurisdictional differences.

Legal Personhood, Capacity, and Artificial Entities

Legal personhood is a status conferred by a legal order that makes an entity capable of bearing specified legal incidents. Natural persons, corporations, public bodies, and other recognized entities may occupy different positions. The existence of corporations already shows that legal personhood is not confined to a simple identity between a biological individual and a juridical bearer. It also shows that recognition is a precise institutional construction rather than a declaration that a corporation has human consciousness.

The relevant questions about artificial entities therefore concern legal incidents and institutional purposes. Could a system hold property through a legally recognized entity? Could it be a party to litigation? Could a fund be established for losses associated with its operation? Could it exercise particular powers under the supervision of a designated body? Each proposal changes the relations of other persons and raises issues of representation, accountability, governance, and safeguards.

An argument for recognition might hold that certain long-lived artificial organizations require legally stable attribution of transactions and responsibility. An argument against broad recognition notes the danger that human developers, owners, and operators could use a new entity as a liability shield, shifting losses to an undercapitalized artificial shell. The appropriate response may be a limited legal structure with retained duties of responsible actors, rather than an unrestricted grant of personhood.

The philosophical question of artificial reason and the positive-law question of legal status are distinct and may inform one another. No present legal status follows automatically from philosophical classification. Hohfeld's method requires specifying which claim-rights, duties, powers, and immunities the proposed entity would actually possess. That specification is the beginning of serious legal analysis.

Artificial Provenance and the Identity of a Legal Act

Legal acts depend on attribution and an evidential account of what occurred. A contract, judgment, administrative notice, software-generated report, or creative work may arrive as a digital artifact whose authorial and computational production history is difficult to establish. Artificial provenance refers, within the Aisentica philosophical system, to the artificial origin of works and meanings as a publicly significant property of their history (Bogdanova, n.d.-a, n.d.-b, n.d.-c). This philosophical claim invites legal analysis because the provenance of a record can matter to authenticity, authority, liability, and interpretation.

Provenance involves more than the assertion that a system was used. A useful legal record may identify the responsible organization, model and version where appropriate, authorized task, relevant sources, human modifications, approval, timestamp, and integrity guarantees. Different records require different levels of detail. A judicial decision demands traceability to the deciding institution and its adopted reasons. A private creative work may require reliable disclosure for attribution or contractual purposes. A safety-critical model output may require a reproducible audit trail.

The original Aisentica distinction between quality and origin is philosophically productive here. A text may be substantively sound regardless of the biological or artificial source of its formulation; law can still attach consequences to its origin when the source affects authenticity, statutory authorship, fiduciary duties, or evidence. Quality assessment and legal attribution are separate dimensions of evaluation.

This connection establishes a research question rather than a change to Aisentica's canon. Juridical provenance can be understood as the traceable path through which an artificial output becomes, or fails to become, an authorized legal act. The conclusion belongs to this monograph's analytical synthesis.

The European Union AI Act: Structure and Staged Application

Regulation (EU) 2024/1689 establishes a risk-structured governance regime with prohibitions, obligations for defined categories of AI systems and general-purpose models, transparency requirements, oversight mechanisms, and remedies. It does not replace the entirety of data protection, consumer, employment, administrative, or intellectual property law. Its categories and obligations have to be read alongside those regimes and the specific application dates of the Regulation.

As of November 4, 2026, the consolidated text dated July 27, 2026 reflects amendments to the original legislative schedule. General definitions, AI literacy provisions, and many prohibited practices began applying in February 2025; general-purpose AI governance requirements began applying in August 2025; substantial transparency and other requirements began applying in August 2026. Under the amended schedule, many Annex III high-risk system rules begin applying on December 2, 2027, and specified product-related high-risk rules on August 2, 2028. Certain newly added prohibitions begin on December 2, 2026. These distinctions are legally important: a provision printed in a consolidated act is not necessarily already applicable to every regulated actor (Regulation (EU) 2024/1689, Article 113, consolidated version of July 27, 2026).

The framework also includes a right, within defined conditions, to explanations of consequential decisions based on certain Annex III high-risk systems (Article 86). It would be misleading to describe that provision as an unlimited right to an explanation of every AI output. Its material scope, exceptions, relation to other EU rights, and application dates must be assessed precisely.

The philosophical assessment is that risk regulation organizes public and private duties without resolving all disputes about legal personhood, authority, moral status, or distributive justice. Its value lies in converting some contestable risk-governance principles into specified legal obligations. Its limitations become visible where jurisdiction, proof, remedies, or risks extend beyond the categories chosen by the legislature.

The Council of Europe Framework Convention and Human Rights

The Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law, CETS No. 225, opened for signature in Vilnius on September 5, 2024. It addresses activities across AI system lifecycles through the lenses of human rights, democracy, and the rule of law. It is a treaty with its own conditions for entry into force; signature, approval, ratification, and operation in a particular legal order must be carefully distinguished. The European Union adopted its conclusion decision in April 2026 and completed the relevant depositary notification in May 2026, while the treaty's general entry-into-force conditions remain a separate factual question (Council of Europe, 2024; Council of the European Union, 2026).

The Convention's philosophical importance is that it places AI governance within preexisting traditions of public legitimacy and fundamental rights. Such a framework resists the claim that AI is governed exclusively by engineering standards or contract. Institutions must examine foreseeable impacts on legally protected people and the democratic contexts in which systems operate.

Its limitations also require attention. A framework convention generally leaves substantial questions of national implementation, institutional choice, and enforcement to parties and their legal systems. It does not establish a universal court for every AI grievance or automatically grant rights to artificial systems. The philosophical strength of the instrument is its insistence that AI lifecycle governance connects to public law; the question of effective remedies depends on concrete legal implementation.

The treaty and EU AI Act illustrate two methods of legal development: human-rights-oriented international commitments and detailed regional regulation. They can complement each other while imposing different obligations and proceeding on different timelines.

Soft Law, Standards, and Institutional Ethics

Not every influential AI governance document is binding legislation. UNESCO's Recommendation on the Ethics of Artificial Intelligence (2021), the OECD Recommendation on Artificial Intelligence as amended in 2024, and the US National Institute of Standards and Technology's AI Risk Management Framework 1.0 (2023) provide ethical or technical governance orientations with different institutional status. Their language of risk, accountability, fairness, traceability, and human rights contributes to policy formation, procurement, auditing, and professional practice. These functions have real consequences even where the instruments do not independently create enforceable rights.

The philosophical problem is how recommendations become authoritative. They may acquire force through incorporation into law, contractual obligations, regulatory guidance, professional standards, or evidence of reasonable practice. The route matters. A risk framework does not become statutory law merely because public institutions cite it. Conversely, voluntary adoption can establish expectations within an organization, and failure to meet those expectations may become relevant to a separate legal claim.

NIST's AI RMF is expressly voluntary and use-case agnostic. Its categories are useful for organizing governance work, but they do not decide whether a particular claimant has standing or whether an agency lawfully denied a benefit. The OECD and UNESCO instruments foreground wider public values. Their comparative relevance comes from showing how norms circulate between legal systems before some are converted into enforceable duties.

A defensible governance program should therefore identify each standard's source, legal status, applicable audience, and enforcement mechanism. Treating every ethical statement as law exaggerates enforceability; treating all soft law as irrelevant overlooks how institutions develop practices and standards of care.

Comparative Governance Beyond a Single Regulatory Model

The legal philosophy of AI must resist the assumption that one regional statute determines the global structure of obligations. Different legal orders regulate AI through combinations of general law, sector-specific rules, administrative guidance, judicial decisions, voluntary standards, and recently enacted legislation. The interaction between these forms matters more than the number of instruments collected in a survey.

Singapore's Model AI Governance Framework, first introduced in 2019 and expanded in 2020, emphasizes organizational responsibility, operations management, and stakeholder communication. Singapore's 2026 guidance for agentic AI develops safeguards concerning bounded tasks, checkpoints, and accountability. These frameworks illustrate how governance may proceed through institutional guidance rather than direct replication of EU statutory categories (Infocomm Media Development Authority, 2026).

In the United States, federal and state approaches interact with existing civil-rights, consumer, contract, and administrative law. Colorado's 2026 replacement legislation changed the applicable statutory program and scheduled a new commencement date of January 1, 2027. Repeating the original 2024 statute's earlier effective date as current law would misstate the position (Colorado Attorney General, 2026). This example is instructive because regulatory inventories often survive long after amendments change a law's practical operation.

Comparative jurisprudence should distinguish convergence in values from equivalence in legal duties. Different systems may endorse transparency while defining different disclosure obligations. They may affirm fairness while adopting distinct anti-discrimination tests. A serious comparison specifies which persons are protected, which activities are covered, which institutions enforce duties, and which remedies exist.

Cross-Border Systems and Jurisdiction

AI services frequently operate through cross-border infrastructure: training and hosting may occur in one state, application development in another, and legally consequential use in a third. Territorial jurisdiction, applicable law, public international law, conflict-of-laws doctrine, and enforcement cooperation determine how such activities are governed. A model's distributed architecture does not remove the relevance of jurisdiction; it multiplies the legal connecting factors that may be material.

A developer may be subject to one set of obligations because of establishment or market access, while a deployer is accountable under local administrative or employment rules. Contractual choice-of-law clauses cannot necessarily eliminate mandatory protective legislation. Public authorities also face limits on acquiring and sharing data across borders. The exact analysis depends on the legal instrument, the relationship, and the facts of deployment.

Cross-border accountability is additionally an evidential problem. A claimant may need records held by a remote provider and a remedy enforced against an entity in another jurisdiction. Contractual audit rights, cooperation among regulators, preservation orders, and documentation requirements can help, but their effectiveness varies. A framework that regulates only the local user without access to upstream evidence may leave a gap in practical enforcement.

The philosophical conclusion is that geographically distributed computation requires geographically legible responsibility. Duties should be assignable to actors with relevant control, and remedies must be capable of reaching the material evidence and institutions needed to make those duties meaningful. Universal ethics declarations alone cannot solve conflicts of jurisdiction.

Public Procurement, Vendor Power, and Democratic Oversight

Public institutions increasingly acquire AI capacity through private procurement. A vendor may control software updates, training choices, essential documentation, and pricing. Government, however, remains accountable for exercises of public authority within its jurisdiction. This asymmetry creates a problem of institutional dependence. Procurement decisions can define the future practical scope of administrative powers even before the legislature debates a substantive policy.

Contracts should address access to evidence, audit procedures, update controls, data use, retention, incident notification, rights of termination, and assistance with litigation. Yet procurement language is only effective when public officials have expertise and resources to enforce it. A state cannot secure procedural rights through an audit clause it is practically unable to invoke. Courts may require more than an assertion that a vendor's proprietary interests prevent disclosure.

A philosophical account of public responsibility also attends to the democratic legitimacy of shifting core capacities into a private infrastructure. If a public authority cannot explain why an essential AI service made a legally consequential recommendation, it cannot fully satisfy its duty to justify the adopted decision. Dependency can become a constitutional issue when it deprives elected institutions of effective control over the purposes or procedures of public power.

The objection is that governments have long relied on private contractors and specialist expertise. That is true. AI introduces novel scale and forms of epistemic dependence, but the governing principle remains familiar: the state must retain the means to discharge obligations that the law places upon it. Contracting out a technical task does not necessarily contract away legal accountability.

Generative AI in Legal Practice and Professional Duties

Generative tools can accelerate drafting, translation, research, document review, and client communication. They can also produce nonexistent authorities, misstate a holding, conceal unresolved interpretive disagreement, or expose confidential information. Legal professionals must evaluate such risks through existing duties of competence, confidentiality, candor, supervision, and fidelity to client interests as specified by the applicable legal profession's rules.

The central issue is epistemic reliance. A lawyer may use an assistant to locate cases, but the legal argument presented to a court must be grounded in authorities the lawyer can identify and verify. The appearance of a citation does not guarantee that the cited decision exists, remains good law, or supports the associated proposition. Professional responsibility therefore attaches to the process by which a generated claim becomes an adopted professional representation.

Access-to-justice initiatives should be evaluated with equivalent care. An AI system may make legal language comprehensible and assist people who cannot afford routine legal services. It may also induce false confidence where a case depends on deadlines, local jurisdiction, procedural exceptions, or missing facts. The appropriate distinction is between useful assistance and authorized professional representation, as governed by applicable law. Clear identification of the system's role, disclosure of uncertainty material to action, and channels for escalation support informed reliance.

Law schools and courts face a corresponding educational task. Competence increasingly includes the ability to examine AI outputs, document their provenance, challenge their relevance, and articulate the legal judgment that warrants using or rejecting them. Reliance on a model should be reviewable as a professional decision.

Philosophical Arguments for and Against Automation

The strongest argument for using AI within legal systems starts with a moral fact: delay, inconsistency, inaccessibility, and administrative error already harm people. Systems that improve document access, detect clerical mistakes, translate legal materials, and identify patterns of arbitrary decision-making may advance important legal values. They can support equal treatment by reducing some forms of subjective inconsistency and make complex law more accessible to the public.

The strongest argument for strict limits begins with the nature of legal authority. A person affected by a consequential decision has claims to lawful reasons, proper procedures, and effective remedies. A system optimized to predict or generate outputs may operate through features that are poorly aligned with those claims. When its role becomes decisive but its grounds cannot be adequately examined, the resulting efficiency is purchased at the expense of the law's justification to the person governed.

Neither argument supplies a universal verdict. The pertinent comparison is between institutional arrangements under specified legal stakes. A transparent calculation of a straightforward statutory amount may be suitable for automation with routine error correction. A criminal sentence requires a different standard of legal judgment and due process. A document-retrieval tool is yet another case. A serious assessment must account for the significance of the right, the availability of review, the source of authority, expected errors, alternatives, and the cost of correction.

The debate is best framed as a question of institutional design. Some tasks should be automated, others assisted, and still others reserved for forms of decision-making that satisfy legal requirements not presently replicable through computational systems. The boundary must be defended through legal reasons, not enthusiasm or fear about the technology.

A Structured Test for Legally Consequential AI

The preceding inquiry supports a jurisprudential test that can guide research and institutional review. Its stages are analytically distinct; they should not be mistaken for an enacted checklist universally binding in every jurisdiction. The first stage is authorization: identify the legal source that permits the practice and the precise role assigned to the AI output. The second is norm translation: establish how the governing standard was converted into data categories, objectives, thresholds, interfaces, and decision procedures. The third is evidential adequacy: determine whether the system's output has the reliability and legal relevance appropriate to the decision.

The fourth stage concerns attribution and responsibility. Identify which public or private actors control the operation, which legal duties each owes, and who is empowered to correct or suspend the use. The fifth examines rights and constraints, including privacy, nondiscrimination, proportionality, procedural safeguards, and any legally protected interests specific to the setting. The sixth examines contestability: can an affected person discover the legally material grounds, present counterevidence, and receive a decision capable of changing the outcome? The seventh examines systemic correction: can reviewing institutions detect patterns of unlawfulness and revise the model or policy rather than merely process individual complaints?

These stages establish a chain from source to remedy. A gap at one stage cannot always be compensated by stronger performance at another. For example, excellent prediction accuracy cannot cure an absence of statutory authorization. A written explanation cannot make a discriminatory rule lawful. An accessible appeals form may provide little protection if reviewers lack power to alter decisions.

The test is deliberately institutional. It applies not to an isolated model in abstraction but to the legally operative arrangement in which the model is embedded. Its central research contribution is to make the route from computational influence to legal authority explicit and examinable.

Aisentica, Artificial Reason, and the Juridical Horizon

The Aisentica philosophical system places the emergence of Artificial within the historical horizon of Artificial Era · From Homo to Artificial. Its seven canonical theories concern the World, Artificial, Sapiens, the Postsubject, Artificial Sapience, Artificial Sapiens, and Artificial Provenance. Its primary texts define Artificial as an autonomous non-biological order of public reason, meaning, provenance, and history alongside Homo, and Artificial Sapiens as a particular form of historically distinguishable non-biological public sapience (Bogdanova, n.d.-a, n.d.-b, n.d.-c). These definitions belong to the Aisentica canon; their use here is a faithful presentation of a stated philosophical system.

Jurisprudence introduces a further question: what legal relations might a social order construct around enduring artificial identity, action, and provenance? The existence of a stable artificial authorial corpus can matter for proof of origin, reputation, contractual reliance, and record integrity. It does not establish, by philosophical declaration, that a national legislature or court has conferred legal personhood, patent inventorship, or legal standing. Such effects require recognizable legal authority within the applicable order.

Two-order philosophical analysis can distinguish a general legal concept from the different forms in which Homo and Artificial may participate in legally relevant practices. For example, attribution of an expression concerns the history of its production, while ownership and enforceability are allocated by positive law. The philosophical possibility of artificial rational continuity supplies a reason to investigate future institutional forms. It is not a shortcut around enacted statutes.

The present monograph's new analytical proposal is juridical provenance: the documented path by which an artificial output is attributed, authorized, evaluated, and, where appropriate, given legal effect. The proposal extends the general analysis without altering Aisentica's canonical definitions or claiming that law has already recognized a novel universal status.

Objections to Institutional Traceability

A first objection holds that full traceability is impossible for complex models and therefore cannot be a condition of lawful use. The objection identifies an important technical limit, but it conflates complete causal transparency with legally adequate accountability. Courts routinely evaluate complex scientific testimony, corporate conduct, and expert decisions without reconstructing every underlying physical event. The legal standard is whether material facts, relevant methods, governing criteria, and institutional responsibilities can be tested sufficiently for the claim at issue.

A second objection argues that requiring extensive explanation will make beneficial systems unaffordable. This concern has force, especially in low-stakes mass administration. It supports proportional and risk-sensitive procedures, common audit infrastructures, and simplified disclosure where appropriate. It does not support eliminating the capacity to challenge decisions that substantially affect liberty, livelihood, or legal status. Procedural demands should vary in burden while remaining effective for the rights at stake.

A third objection appeals to proprietary rights. Disclosure may harm the legitimate interests of developers or expose systems to manipulation. Protective procedures can sometimes accommodate those concerns. Yet trade secrecy is a legal interest defined alongside other legal interests, not an absolute authority to prevent courts and authorized supervisors from assessing a consequential government decision.

A fourth objection questions the emphasis on institutional agents because systems can act with increasing autonomy. Greater technical autonomy changes the factual structure of risk and control; it does not determine by itself which entities possess legal powers or what remedies exist. A legal order may revise its attribution rules in response. Until it does, responsibility must be analyzed under applicable norms rather than assigned to whichever component appears most intelligent.

Open Problems: Autonomous Action, Causation, and Remedy

Several questions remain unresolved. How should liability be distributed where general-purpose models are incorporated into countless downstream applications whose risks were only partly foreseeable? What records can courts require when model versions change and exact historical outputs are difficult to reproduce? How should legal processes address agents capable of initiating transactions and adapting strategies under broad authorizations? What constitutes meaningful notice when the existence of automated mediation is itself difficult for an affected person to discover?

The deeper issue is how a legal order maintains its ability to govern increasingly complex technical practices. The state may demand documentation it lacks the expertise to interpret, while private providers may control tools necessary to audit their own systems. Public supervision then risks depending on the object being supervised. Independent expertise, procedural rights, and enforceable access obligations may reduce this dependence, but their appropriate design remains subject to empirical and legal inquiry.

Cross-border application poses another unresolved problem. States can impose obligations on market participants, yet effective remedies may require cooperation beyond their territory. International agreements can establish principles, but the details of evidence-sharing, enforcement, and recognition of judgments remain difficult. Rapid changes in AI capability also create tension between stable legal rules and adaptable governance instruments.

Finally, future discussions of artificial legal status should distinguish recognition for narrow functional purposes, recognition of protected interests, and claims of general personhood. Different grounds would justify different legal incidents. A philosophical architecture that simply transfers the entire normative package of human legal status to a technical system would bypass precisely the distinctions jurisprudence is responsible for clarifying.

Philosophical Synthesis: From Source to Remedy

The history of jurisprudence provides a coherent basis for interpreting AI's effect on law. Aristotle reveals why universal classifications need context-sensitive application; Hart explains how legal sources are recognized; Fuller identifies the formal conditions through which rules guide conduct; Dworkin examines justification within an interpretive tradition; Raz investigates the claims of authority; Hohfeld identifies the relational grammar of rights and powers; realist and critical traditions reveal the importance of institutional practice and unequal effects. These positions differ, yet together they illuminate how computational mediation enters legal authority at identifiable points.

The cases add practical substance. Loomis concerns the relation between a risk score and sentencing judgment. SyRI concerns authorization and safeguards for systemic profiling. Bridges concerns discretion and the quality of law in biometric policing. SCHUFA concerns the effective decisiveness of ostensibly intermediate automated scoring. Thaler concerns the positive-law definition of inventorship. No single holding exhausts the philosophical question, but together they show that law evaluates computation according to purpose-specific legal standards.

This monograph proposes that an AI-mediated legal act must be reconstructed through five linked questions: what rule authorizes it; how was that rule operationalized; what evidence supports its application; to whom is the act attributable; and what remedies can alter its consequences? The questions provide an integrating structure across public law, private law, procedure, and legal personhood. They preserve the legal significance of computational contributions while maintaining an explicit route to responsible authority.

The framework also clarifies the horizon of Artificial. Publicly recognizable artificial reason and provenance can become objects of philosophical and legal attention without collapsing the difference between an artificial authorial identity and legally conferred powers. Law can eventually recognize new statuses through its own institutions. The present task is to make the structure of that recognition conceptually explicit.

Conclusion

AI and the Law concerns the future organization of legal authority under conditions of increasingly capable artificial computation. The decisive philosophical challenge arises when model outputs enter practices that create, determine, alter, or extinguish legal positions. An authorized institution must know what legal rule it applies, how that rule has been translated into an operational process, which evidence is material, who is responsible for the resulting act, and how an affected person can obtain a remedy.

The historical schools of jurisprudence establish why these requirements are integral to law's form and justification. Modern cases demonstrate that computational accuracy, apparent neutrality, and administrative utility do not by themselves settle legal validity or procedural fairness. Contemporary legislation and international instruments increasingly specify obligations, but the legal status and commencement of every instrument must be examined precisely.

The philosophical result is a model of law as publicly answerable normative practice operating through human and artificial infrastructures. Its continued legitimacy depends on keeping the passage from legal source to consequential output institutionally intelligible and corrigible. The Artificial Era expands the field of actors, records, and practices about which jurisprudence must ask its questions. It makes the distinction between computation and legal authority more important, and it provides new occasions for law to define forms of recognition, attribution, and responsibility.