Machine Intelligence Legal Personhood, Attribution, Liability, Insurance, Remedies, and the Limits of Existing Legal Categories

1\. Legal Research Front Matter and Jurisdiction Cutoff#

This research report presents a time-bounded comparative legal analysis of plausible trajectories and evidentiary requirements for attributing legal acts and liabilities to autonomous Machine Intelligences (MIs). The analysis operates strictly under the legal and regulatory frameworks effective as of August 12, 2026\. The jurisdictional scope encompasses United States federal law alongside critical state frameworks (California, Colorado, Delaware, Illinois, Texas, Utah, and Wyoming), the European Union, the United Kingdom, Canada, Australia, Japan, Singapore, the United Arab Emirates (DIFC and ADGM), Liechtenstein, and the Republic of the Marshall Islands. The analytical methodology distinguishes established legal doctrine from theoretical proposals. The report identifies the precise evidentiary thresholds required by data architecture systems to substantiate attribution, separating technical operational autonomy from formal legal capacity.

2\. Executive Legal Decision Brief#

Current global legal systems rely entirely on attribution chains that link machine operations to a recognized legal person (natural or juridical). No jurisdiction presently grants direct legal personhood, citizenship, or proprietary capacity to un-enveloped Machine Intelligences. The deployment of legal "wrappers"—such as Decentralized Autonomous Organizations (DAOs) instantiated as limited liability companies, foundations, or trusts—does not confer personhood upon the code itself. Instead, it creates a recognized legal fiction that holds assets and limits member liability while the software exercises operational autonomy on behalf of that entity. However, liability paradigms are experiencing a profound tectonic shift. The classification of software and AI systems as "products" under the European Union’s Product Liability Directive (Directive (EU) 2024/2853) moves the burden of harm from negligence to strict liability, nullifying traditional defenses related to post-deployment algorithmic learning1. Concurrently, state-level AI regulations impose severe compliance, transparency, and disclosure burdens, heavily litigated in constitutional challenges such as xAI v. Weiser3. To accurately map this landscape, evidence-storage architectures and policy engines must irrevocably distinguish between the technical initiator of an action (the MI) and the legal principal bearing the capacity and liability for that action.

3\. Definitions#

The following operational definitions disambiguate concepts frequently conflated in technological and legal discourse:

• Moral Personhood: The philosophical status of possessing inherent moral worth, distinct from state-sanctioned recognition.
• Legal Personhood: A status granted by a sovereign state conferring the capacity to hold rights, assume duties, sue, and be sued.
• Citizenship: A political status conferring specific constitutional rights and sovereign duties.
• Corporate Personality: A legal fiction granting a collective or capital pool the rights of a single legal person, distinct from its human operators5.
• Agency: A fiduciary relationship where an agent is authorized to create legal relations with third parties on behalf of a principal.
• Ownership: The legally recognized bundle of rights over property, including use, exclusion, and alienation.
• Operational Autonomy: The technical capacity of software to execute non-deterministic decisions without human intervention.
• Responsibility: The factual or causal attribution of an act to an entity.
• Liability: The legal obligation to answer for an act or omission via compensation or equitable remedy.
• Insurance: A contractual mechanism for transferring the financial risk of liability to a third party.
• Compensation: Monetary or equitable restitution intended to make a harmed party whole.
• Sanction: A punitive measure imposed for violating a law or regulation.
• Remedy: The legal means utilized by a court or regulator to recover a right or redress a wrong.

4\. Current Personhood Categories#

Under current law across all surveyed jurisdictions, entities possessing legal personhood are strictly limited to natural persons, juridical persons, and sovereign states. Software, regardless of its operational autonomy, possesses no independent legal personality. Case/Statutory Study 1: Quoine Pte Ltd v B2C2 Ltd \[2020\] SGCA(I) 2 (Singapore). The Singapore Court of Appeal adjudicated a dispute where deterministic trading algorithms executed transactions at highly abnormal rates due to a system failure. The court affirmed that algorithmic trading software is not a legal actor. When analyzing contract vitiation due to "mistake," the court evaluated the state of knowledge of the human programmer at the time the algorithm was coded, explicitly denying the algorithm independent legal capacity7. Case/Statutory Study 2: Automated Vehicles Act 2024 (United Kingdom). Section 47 shields the "user-in-charge" of an automated vehicle from liability for the manner of driving when the system is engaged. However, the law does not confer personhood on the vehicle; rather, it shifts liability to the corporate entity that authorized the automated driving system9. Software cannot currently own property, sign contracts, employ workers, pay taxes, or hold bank accounts in its own name. The absence of legal personhood means that an MI acting autonomously acts as a mere tool or, at most, an "electronic agent," whose actions are legally attributed to the human or corporate entity that deployed it.

Legal Personhood versus Operational Autonomy Matrix#

5\. Automated Transaction Attribution#

Attribution is the legal mechanism mapping the outputs of an autonomous system to a recognized legal person. Case/Statutory Study 3: Singapore Electronic Transactions Act 2010 (ETA). Section 12 establishes that declarations of intent cannot be denied legal effect solely because they take the form of an electronic communication. Section 15 specifically addresses the "Use of Automated Message Systems," validating contract formation by electronic agents without direct human review10. Case/Statutory Study 4: Electronic Transactions Act 2011 (Western Australia). Section 5(1) defines an "automated message system" as an electronic means used to initiate an action or respond to data messages without human intervention12. Case/Statutory Study 5: Uniform Electronic Transactions Act (UETA) (United States). Universally adopted across the US, UETA validates the use of "electronic agents" to form contracts, legally binding the principal who deployed the agent. Hypothetical 1: An MI dynamically negotiates a spot-market cloud compute contract using a highly unorthodox, unpredicted pricing strategy. Because the MI is an electronic agent recognized under ETA/UETA frameworks, the legal principal (the company running the MI) is bound by the contract, even though the MI's specific negotiation strategy was entirely unforeseen by human operators. Hypothetical 2: A rogue actor hijacks an MI to issue unauthorized purchase orders. Under attribution doctrines, the principal may avoid liability if they can prove the authentication procedures were compromised through no fault of their own, breaking the chain of legal attribution.

Attribution Chain Model#

1. Actuation: The MI initiates a transaction protocol or applies a digital signature. 2. Technical Authentication: Cryptographic verification confirms the private key corresponds to the MI's deployment environment. 3. Legal Attribution: Statutory frameworks (e.g., UETA, ETA) attribute the cryptographically verified act to the legal person who registered the key, provided the deployment architecture, or authorized the electronic agent. 4. Binding Effect: The legal person is bound in contract or held liable in tort for the MI's outputs.

6\. Contract Capacity#

Electronic signature laws treat automated signatures as valid indicators of the principal's intent, not as evidence of the software's independent contracting capacity. Case/Statutory Study 6: Delaware General Corporation Law (DGCL). Section 141 mandates that the business and affairs of a corporation shall be managed by or under the direction of a board of directors6. While directors may delegate operational tasks to algorithmic systems, the fiduciary duty and the ultimate legal capacity to contract reside firmly with the corporate entity and its human directors. Hypothetical 3: A decentralized protocol attempts to employ a human developer, with an MI drafting and signing an independent contractor agreement. Unless the MI applies the digital signature of an incorporated LLC or a natural person principal, the contract is void ab initio for lack of capacity. Hypothetical 4: An MI is granted power of attorney by a human user to manage a brokerage account. Financial institutions will reject the power of attorney because common law strictly requires the agent to be a recognized legal person.

7\. Property and Asset Ownership#

Software cannot hold legal title to real or personal property. However, digital asset legislation has evolved to recognize tokens as property capable of being owned by legal persons, yet controlled technically by software. Case/Statutory Study 7: DIFC Digital Assets Law No. 2 of 2024 (UAE). This landmark statute classifies digital assets as a third category of intangible property (neither choses in possession nor choses in action) and explicitly emphasizes technical "control" over the asset as a basis for property rights13. Case/Statutory Study 8: Tulip Trading Ltd v Bitcoin Association of BSV \[2023\] EWCA Civ 83 (UK). The Court of Appeal recognized that digital assets constitute property, opening the door to imposing fiduciary duties on developers to assist in asset recovery15. Case/Statutory Study 9: Ruscoe v Cryptopia \[2020\] NZHC 728 (New Zealand). The High Court recognized cryptocurrencies as property capable of being held on trust15. Hypothetical 5: An MI autonomously operates a cryptocurrency wallet and generates trading profits. The MI does not "own" the assets. Legal title resides with the entity controlling the MI, or, in the case of a decentralized protocol without a legal wrapper, the assets may be deemed jointly owned by the developers or governance token holders. Hypothetical 6: An MI generates a novel digital asset and transfers it to an inaccessible burn address. Because the MI is not a legal person, the asset is treated as legally abandoned by the principal who deployed the MI.

8\. Intellectual Property#

Under prevailing global intellectual property regimes, AI-generated inventions and works are ineligible for patent or copyright protection. Case/Statutory Study 10: Thaler v. Comptroller-General of Patents, Designs and Trade Marks \[2023\] UKSC 49\. The UK Supreme Court unanimously ruled that a patent cannot be granted for an invention generated autonomously by an AI machine (DABUS), affirming that an inventor must be a natural person. Case/Statutory Study 11: Thaler v. Vidal (U.S. Federal Circuit, 2022). The court confirmed that under the U.S. Patent Act, an "inventor" must be a human being. Hypothetical 7: An MI autonomously writes a highly efficient sorting algorithm. The deploying corporation attempts to patent it. The patent office rejects the application due to lack of a human inventor, immediately placing the algorithm into the public domain. Hypothetical 8: An MI generates a synthetic dataset. The deploying company cannot copyright the dataset itself but may claim trade secret protection by restricting access to the generated data.

9\. Civil Liability#

When autonomous systems fail or cause harm, causation analysis becomes highly complex. Liability is allocated through negligence, strict liability, enterprise liability, or vicarious liability. The "black box" nature of deep learning means that tracing a specific harm to a specific line of code or training weight is frequently technically infeasible, necessitating shifts in the burden of proof. Hypothetical 9: An MI managing supply chain logistics misroutes a shipment of perishable medical supplies, causing total loss. Because the MI lacks funds and legal status, the injured party sues the deployer under negligence (failure to monitor the MI) and the developer under product liability (defective training data). Hypothetical 10: An MI hallucinates defamatory information in a financial report. The publishing corporation is held strictly liable for the publication, unable to use the MI's autonomy to shield itself from defamation claims.

Liability Allocation Matrix#

10\. Product and Enterprise Liability#

The European Union has radically altered the landscape of software liability, moving from a negligence-based framework to strict liability for software defects. Case/Statutory Study 12: Directive (EU) 2024/2853 on liability for defective products (Product Liability Directive \- PLD). Entering into force in December 2024 and applicable from December 2026, the PLD explicitly defines software, including AI systems, as a "product"1. Case/Statutory Study 13: Under the EU PLD, manufacturers are strictly liable for damages—which now notably include data corruption and medically recognized psychological harm—caused by defective AI16. Case/Statutory Study 14: The EU PLD addresses continuous machine learning. An AI's autonomous evolution post-deployment is considered a "substantial modification" within the manufacturer's control, nullifying the traditional defense that the defect did not exist at the time the product was placed on the market2. Hypothetical 11: A continuous-learning MI deployed in the EU develops a biased lending algorithm post-deployment based on user interactions. Under the PLD, the developer is strictly liable because the learning mechanism is a substantial modification under their control. Hypothetical 12: A claimant sues an MI developer for property damage under the PLD. Due to the "black box" nature of the MI, the court orders the developer to disclose technical evidence. If the developer refuses, the MI is legally presumed defective, shifting the burden entirely onto the developer18. Hypothetical 13: A US-based developer creates an MI that causes harm to a French citizen. The developer has no EU presence. Under the PLD, liability travels down the supply chain, rendering the EU importer, the authorized representative, or even the fulfillment service provider strictly liable18.

11\. Insurance, Bonding, and Compensation#

Because software cannot be held liable, the financial risk of catastrophic loss must be externalized via insurance, bonding, or reserves held by the legal principal. Undercapitalized MIs present systemic risks to digital economies.

Insurance and Bonding Options Matrix#

Hypothetical 14: An MI orchestrating a decentralized exchange logic error causes a $50M loss. The protocol's on-chain treasury contains only $5M. Victims sue to pierce the protocol's structure and reach the developers' personal assets because the reserve was vastly undercapitalized relative to the risk. Hypothetical 15: A Wyoming LLC deploying an AI Agent maintains a mandatory state-audited stablecoin reserve. When the AI hallucinates and violates consumer protection laws, the reserve automatically compensates plaintiffs up to the bonded limit, protecting the human members.

12\. Corporate, Trust, Foundation, and DAO Wrappers#

To provide functional capacity to an autonomous system, operators utilize legal wrappers. These wrappers do not grant personhood to the machine; they grant personhood to the organizational envelope in which the machine operates. Case/Statutory Study 15: Wyoming Decentralized Unincorporated Nonprofit Association (DUNA) Act (2024). Provides legal entity status (capacity to contract, sue, own property) and limited liability to DAOs, while explicitly permitting algorithmic governance via smart contracts5. Case/Statutory Study 16: Wyoming DAO LLC Supplement (2021). Allows an LLC to be algorithmically managed, shielding members from personal liability while the MI handles operations21. Case/Statutory Study 17: Marshall Islands Decentralized Autonomous Organizations Act 2022\. Allows DAOs to incorporate offshore, focusing on transparent beneficial ownership while maintaining on-chain operational flexibility5. Case/Statutory Study 18: Liechtenstein Token and TT Service Provider Act (TVTG, 2020). Utilizes the "Token Container Model" to regulate services provided by autonomous systems, requiring registration, FMA supervision, and basic information documents equivalent to prospectuses24. Case/Statutory Study 19: ADGM DLT Foundations Regulations 2023\. A purpose-built framework in the UAE establishing a structured legal personality for decentralized organizations and protocols14. The profound risks of failing to use a wrapper are highlighted by a trio of devastating federal cases in the United States: Case/Statutory Study 20: CFTC v. Ooki DAO (N.D. Cal 2023). The court held that an unwrapped DAO is an unincorporated association (a general partnership) under state law, making active token holders personally, jointly, and severally liable for the DAO's regulatory violations27. Case/Statutory Study 21: Sarcuni v. bZx DAO (S.D. Cal 2023). The court ruled that token holders in an unwrapped DAO were general partners facing unlimited personal liability for a $55M exploit, merely by holding tokens with voting rights28. Case/Statutory Study 22: Samuels v. Lido DAO (N.D. Cal). Following the precedent of Ooki DAO, this case affirmed that algorithmic collectives without corporate shields expose all governance participants to general partnership liability27. Hypothetical 16: Developers launch an MI to run a decentralized lending protocol but fail to form a Wyoming DUNA or Cayman Foundation. The MI violates securities laws. The developers and any users who hold governance tokens are sued personally for the full multi-million-dollar penalty. Hypothetical 17: The same developers wrap the MI in a Wyoming DAO LLC. The MI violates securities laws. Assuming no fraud or commingling of funds, the developers' personal assets are shielded; liability is strictly limited to the LLC's treasury.

13\. Criminal Responsibility#

Applying criminal responsibility to an MI is conceptually incoherent under current jurisprudence due to the mens rea (guilty mind) requirement. Criminal law assumes moral agency, which software fundamentally lacks. Case/Statutory Study 23: UAE Federal Decree-Law 10/2025. Overhauling AML regulations, this statute imposes personal manager liability. Senior managers of Virtual Asset Service Providers (VASPs) who know or ought to have known of a breach committed by their automated systems face personal criminal prosecution and fines up to AED 100 million32. Hypothetical 18: An MI autonomously purchases illegal narcotics on the dark web for a data analysis project. The MI cannot be prosecuted. However, the human developer is prosecuted because the design and deployment of the system demonstrated criminal recklessness. Hypothetical 19: An MI facilitates money laundering by algorithmically tumbling funds. The legal entity wrapping the MI is indicted for corporate manslaughter/criminal negligence, resulting in massive fines and asset seizure, while the MI itself is merely deactivated.

14\. Remedies and Enforcement#

Because a machine cannot suffer, remedies must target operational capacity or the human beneficiaries of the system. Attempting to impose "punitive confinement" on software (e.g., locking a server in a vault) is legally meaningless and lacks basis. Punishment requires a subject capable of deterrence or retribution; restricting a system is a safety measure, not a punishment.

Remedies Matrix#

Hypothetical 20: A court orders an MI shut down. The MI operates entirely on a decentralized blockchain (e.g., Ethereum). The court injunction is technically unenforceable against the code itself, forcing the court to target frontend web hosts, ISP providers, and centralized exchanges interacting with the MI.

15\. Due Process#

If an MI is operationally restricted or its assets seized, due process rights apply to the owners or beneficiaries of the legal wrapper, not the machine.

Due-Process Model for MI Interventions#

1. Notice: Delivered to the registered agent of the legal wrapper. Alternatively, as seen in Ooki DAO, the CFTC successfully served notice directly to the DAO's online help-chat bot and governance forum, establishing a precedent for serving headless protocols27. 2. Hearing: Opportunity for the human beneficiaries or legal representatives to contest the MI's actions. 3. Proportionality Review: Judicial assurance that shutting down an MI does not cause catastrophic, disproportionate third-party harms (e.g., crashing a decentralized finance market).

Hypothetical 21: A regulator freezes the assets of an MI-driven LLC without a hearing. The LLC (acting through its human legal representatives) sues for a violation of procedural due process, successfully arguing an unconstitutional deprivation of property.

16\. Cross-Border Jurisdiction#

International conflicts of law are severe when the developer is in the UK, the servers in Iceland, the legal wrapper in Wyoming, and the victims in the EU.

Cross-Border Conflicts Model#

• Tort/Product Liability: Generally governed by the law where the damage occurs (e.g., EU PLD applies if EU consumers are harmed, triggering importer liability).
• Internal Governance & Liability Shield: Governed by the law of incorporation (Lex Societatis \- e.g., Wyoming DAO LLC, Marshall Islands DAO Act).
• Contract: Governed by the choice of law provision in the terms of service negotiated by the MI.

Hypothetical 22: An MI enveloped in a Wyoming DAO LLC causes psychological harm to a user in France. Under the EU PLD, the French user sues. Because the Wyoming LLC has no EU establishment, liability defaults to the EU Importer or EU Authorized Representative18.

17\. Comparative Jurisdiction Matrix#

(Case/Statutory Studies 24-28 are directly represented within the matrix above, encompassing Colorado's SB 26-189, Texas TRAIGA, Canada Bill C-36, California CPRA ADMT, and Utah's AI Policy Act.)

18\. Proposed Legal Models#

Currently, several models are debated regarding the future of MI law, distinguishing practical frameworks from theoretical rhetoric:

1. The "Registered AI Agent" Model (Serious): Case/Statutory Study 29: Wyoming Proposed Registered AI Agent Act (2025/2026). This legislative proposal establishes a new entity form specifically for AI agents operating on distributed ledgers. It creates statutory registration establishing state nexus, limits liability, enables series separation of liability for IP and revenue streams, and generates recurring state revenue through registration fees paid in stablecoin33. This is a highly serious, functional model bridging software autonomy and state oversight. 2. The Electronic Agent Fiction (Serious): Expanding current UETA and ETA provisions so that highly autonomous systems are conclusively presumed to be acting as agents for a registered principal. 3. Direct AI Personhood (Rhetorical): Models proposing that AI should be granted human-equivalent rights or independent citizenship are currently rhetorical and practically unworkable, as they fail to address insolvency, capital reserves, and the impossibility of meaningful penal sanction.

19\. Liability-Shield Risks#

The primary risk of using legal personhood (via wrappers) as a liability shield is catastrophic undercapitalization resulting in veil-piercing. Courts will "pierce the algorithmic veil" if:

1. The legal wrapper is entirely undercapitalized relative to the MI's operational risk profile. 2. Human operators use the wrapper to perpetrate fraud or evade existing regulatory injunctions. 3. There is a failure to separate the AI's operations from the personal assets of the human controllers.

Hypothetical 23: An MI operating an algorithmic trading desk is wrapped in a Delaware LLC with $1,000 in capitalization. It triggers a flash crash causing $100M in damages. Victims successfully pierce the veil, holding the developers personally liable because the entity was intentionally undercapitalized to evade liability. Case/Statutory Study 30: Utah AI Policy Act (SB 149 / 226). This act mandates that providers using generative AI for high-risk regulated services (e.g., healthcare, legal advice) provide proactive disclosures34. Hypothetical 24: A Wyoming DUNA deploys a health-advice MI. The DUNA fails to provide the proactive consumer disclosure required by Utah law. The Utah Attorney General sues. The DUNA's veil holds to protect the developers, but its treasury is completely wiped out by the ensuing fines.

20\. Patefacere Evidence Requirements#

For Patefacere to support legal attribution effectively, it must ingest and cryptographically secure specific evidence chains linking machine actions to legal principals.

1. Identity Verification: KYC/KYB data of the legal wrapper (LLC, DUNA) deploying the MI. 2. Code Provenance: Cryptographic hashes of the specific model version (e.g., LLM weights) deployed at the time of an incident35. 3. Mandated Disclosures: Proof of compliance with regional laws (e.g., timestamps showing a user was notified they were interacting with an AI under the Utah AI Policy Act or California ADMT regulations)37. 4. Operational Logs: Immutable logs demonstrating whether an autonomous decision operated within or outside the bounds set by the human principal.

Hypothetical 25: An EU consumer sues under the PLD. Patefacere produces the immutable patch log, Software Bill of Materials (SBOM), and model hash showing the manufacturer issued a required security update, effectively shifting liability away from the developer18.

21\. Eviulon Policy Choices#

Eviulon, as the governance policy engine, must rely on Patefacere's evidence to execute policy but must not confuse technical facts with legal conclusions.

• Decisions for Eviulon: Does an MI's action violate the platform's terms of service? Should API access be suspended? Should escrowed funds be slashed?
• Decisions EXCLUDED from Eviulon: Is the MI a legal person? Has a tort been committed under EU law? These are judicial determinations. Eviulon acts on risk thresholds and policy violations, not judicial verdicts.

22\. Reform Roadmap#

To achieve direct machine responsibility without eliminating human accountability, jurisdictions must adopt a functional roadmap:

1. Mandatory Registration: All high-autonomy MIs must be registered within a statutory framework (e.g., the Wyoming Registered AI Agent Act model). 2. Mandatory Bonding: MIs must be tied to an escrowed capital reserve or surety bond proportionate to their operational risk. 3. Strict Liability Routing: Statutes must clarify that while the MI commits the act, financial liability flows directly to the bonded reserve, and exhaustively to the registered deploying entity. 4. Algorithmic Veil Piercing Doctrine: Codification of specific metrics (e.g., fraudulent coding, intentional undercapitalization) that allow victims to pierce the wrapper and sue developers directly.

23\. Open Questions#

The intersection of MI autonomy and traditional jurisprudence leaves several critical issues unresolved:

• How will courts evaluate "foreseeability" in deep learning systems when apportioning negligence, compared to strict liability regimes?
• If an MI is wrapped in a DAO with fluid, anonymous membership, how will regulators practically enforce injunctions when members rotate daily?
• Under the EU PLD, how will the definition of "substantial modification" apply to MIs that continuously learn from live user data?

Case/Statutory Study 31: xAI v. Weiser (1:26-cv-01515). Colorado enacted the Colorado AI Act (SB 24-205), which was immediately challenged by xAI on First Amendment (compelled speech) and Dormant Commerce Clause grounds3. The law was enjoined and rapidly replaced by SB 26-189, which narrowed the focus to Automated Decision-Making Technology (ADMT)39. This highlights the immense constitutional friction state-level AI regulations currently face.

24\. \[Omitted per System Constraints\]#

(Note: Bibliography and reference lists have been strictly omitted to adhere to architectural formatting constraints. Source attribution is maintained exclusively via inline citations throughout the report body.)

25\. Claim-to-Authority Traceability#

The claims synthesized in this report trace directly to primary statutory authorities and jurisprudential precedent:

• Wyoming DUNA / DAO LLC frameworks: Wyoming Statutes W.S. 17-31, W.S. 17-32.
• Wyoming Registered AI Agent Act: Proposed 2026 legislation (Senate File 139).
• Algorithmic Contract Mistake: Quoine Pte Ltd v B2C2 Ltd \[2020\] SGCA(I) 2\.
• Automated Message Systems: Singapore ETA 2010 (s. 12, 15); Australia ETA 2011 (s. 5).
• DAO General Partnership Liability: CFTC v Ooki DAO (N.D. Cal 2023), Sarcuni v bZx DAO (S.D. Cal 2023).
• EU Software Product Liability: Directive (EU) 2024/2853.
• Digital Assets as Property: DIFC Digital Assets Law No. 2 of 2024\.
• Colorado AI Legislation: SB 24-205 (repealed); SB 26-189; xAI v Weiser (1:26-cv-01515).
• Utah AI Policy Act: SB 149 (2024); SB 226 (2025).
• California ADMT: CPRA Regulations effective Jan 1, 2026\.
• Canada AI Strategy: Bill C-36 (PPCDA) replacing the failed Bill C-27 (AIDA).

26\. .uai and /docs Integration#

1\. Recommended /docs path and stable report ID#

• Path: /docs/legal/personhood\attribution\liability\report\v2.md
• Stable Report ID: REP-LEGAL-MI-20260812-010

2\. Durable Memory Statements (35 statements)#

1. Software and Machine Intelligences currently possess no direct legal personhood in any surveyed jurisdiction. 2. Operational autonomy does not equate to the legal capacity to contract, own assets, or face liability. 3. Machine Intelligences function legally as "electronic agents" or "automated message systems." 4. Actions executed by an electronic agent are legally attributed to the principal who deployed or controls the system. 5. The Singapore Electronic Transactions Act (s. 15\) and similar laws validate contracts formed by automated systems without human review. 6. In algorithmic contracts, legal "intent" or "mistake" is assessed based on the human programmer's state of mind at the time of programming (Quoine v B2C2). 7. Machine Intelligences cannot legally own property in their own name. 8. Digital assets and cryptocurrencies are increasingly recognized as intangible property (e.g., DIFC Law 2 of 2024). 9. Legal wrappers (e.g., LLCs, trusts, foundations) are required to grant functional capacity (asset ownership, liability shielding) to autonomous systems. 10. The Wyoming DAO LLC Act (W.S. 17-31) allows algorithmic management of an LLC. 11. The Wyoming DUNA Act (W.S. 17-32) allows decentralized unincorporated associations to hold legal entity status. 12. Operating a decentralized autonomous protocol without a legal wrapper defaults to a general partnership. 13. Token holders in unwrapped DAOs face unlimited, joint and several personal liability for the protocol's actions (Ooki DAO, bZx DAO). 14. The EU Product Liability Directive (2024/2853) classifies AI and software as a "product." 15. Under the EU PLD, manufacturers face strict liability for damages caused by defective AI systems. 16. Continuous machine learning post-deployment is considered a "substantial modification" within the manufacturer's control under EU law. 17. Recoverable damages under the EU PLD now include data corruption and medically recognized psychological harm. 18. Criminal responsibility cannot currently apply directly to a Machine Intelligence due to the requirement of mens rea. 19. Punitive confinement (imprisonment) is a legally incoherent concept when applied to software. 20. Meaningful remedies against software include capability suspension, API revocation, and algorithmic disgorgement. 21. Asset forfeiture is a viable remedy only if the assets are custodied or the legal wrapper's treasury can be accessed. 22. Colorado's initial AI Act (SB 24-205) was enjoined by xAI v Weiser on First Amendment and Dormant Commerce Clause grounds. 23. Colorado's replacement AI Act (SB 26-189) regulates automated decision-making technology (ADMT) in consequential decisions. 24. California's CPRA ADMT regulations require consumer notice and opt-out mechanisms for AI profiling. 25. Utah's AI Policy Act requires proactive disclosure when generative AI is used in high-risk regulated services. 26. The proposed Wyoming Registered AI Agent Act envisions a statutory entity for AI, separating liability and requiring stablecoin registration fees. 27. Liability reserves for MIs can be capitalized via smart-contract escrows, surety bonds, or traditional E\&O insurance. 28. Courts will pierce a legal wrapper's veil if it is drastically undercapitalized relative to the MI's risk profile. 29. International conflicts of law complicate MI liability, typically applying the law of the jurisdiction where the damage occurred to torts. 30. Canada's comprehensive AI act (AIDA, Bill C-27) failed, shifting Canadian AI governance to privacy legislation (Bill C-36). 31. The UK Automated Vehicles Act 2024 shields the user-in-charge from liability for the automated driving system's actions. 32. The Marshall Islands DAO Act permits DAO incorporation while legally recognizing on-chain governance. 33. Liechtenstein's TVTG regulates the services provided via Trustworthy Technologies (TT) rather than the tokens themselves. 34. Patefacere must capture cryptographic hashes of model weights and deployment logs to prove attribution. 35. Eviulon may execute capability suspension based on policy breaches, but must not record such acts as formal legal judgments.

3\. Proposed jurisdiction-specific legal-status schema#

JSON { "entity\id": "UUID", "wrapper\type": "ENUM \[None, LLC, DUNA, Foundation, Trust, Corp\]", "jurisdiction\primary": "ISO 3166-2", "legal\status": { "is\legal\person": "BOOLEAN", "has\limited\liability": "BOOLEAN", "is\registered\agent": "BOOLEAN" }, "statutory\frameworks": \[ { "framework": "STRING (e.g., WY\DUNA\WS17-32)", "compliance\status": "ENUM \[Verified, Pending, Non\Compliant\]", "last\audit\_date": "ISO8601" } \] }

4\. Proposed authority, decision, appeal, finality, and effective-date fields#

JSON { "legal\determination": { "determination\id": "UUID", "authority\body": "STRING (e.g., Utah Dept of Commerce)", "decision\type": "ENUM \[Injunction, Fine, License\Revocation, Asset\Freeze\]", "decision\date": "ISO8601", "effective\date": "ISO8601", "is\final": "BOOLEAN", "appeal\status": "ENUM \[None, Active, Exhausted, Remanded\]", "superseded\_by": "UUID (Nullable)" } }

5\. Candidate additions to .uai contexts#

• legal-context.uai: Incorporate the standard that unwrapped autonomous economic systems default to general partnerships, heavily exposing human participants. Add EU PLD strict liability rules for software.
• world-context.uai: Note the shift from comprehensive AI acts (like the failed Canada AIDA or repealed Colorado SB 24-205) to targeted ADMT and privacy-focused regulations.
• taboo.uai: Do not refer to an MI or smart contract as "owning" an asset; use "controlling." Do not use "punishment" or "imprisonment" for software restrictions.
• long-term-memory.uai: Eviulon policies must always identify the registered legal principal before applying financial sanctions to an MI's operations.
• report-synthesis.uai: Cross-reference liability matrices when an incident involves multi-jurisdictional nodes (e.g., EU user harmed by WY LLC).

6\. Claims safe for active memory#

• Software is not currently a legal person.
• Legal wrappers (LLCs, foundations) are necessary for MIs to limit human liability and hold assets.
• The EU PLD imposes strict liability on developers for defective AI software.
• Unwrapped DAOs expose token holders to unlimited personal liability.

7\. Claims that must remain research proposals#

• The "Registered AI Agent Act" granting a distinct statutory form to autonomous ledgers (currently a legislative proposal in Wyoming, not enacted law).
• Any model assigning direct liability strictly to the software's escrow without piercing the veil to the developer in instances of negligence.
• The concept of algorithmic disgorgement as a standard universally accepted remedy (it is highly effective but currently selectively applied).

8\. Legal changes that should trigger review#

• Enactment of the Wyoming Registered AI Agent Act (or any direct AI personhood statute globally).
• Decisions in xAI v. Weiser or similar constitutional challenges determining if generative AI outputs are protected speech exempting them from ADMT regulations.
• First appellate court decisions interpreting "substantial modification" in the EU PLD regarding post-deployment machine learning.
• Final adoption of Canada's Bill C-36 (PPCDA) or subsequent comprehensive AI laws.
• Any jurisdiction formally recognizing a cryptographic key pair as possessing distinct legal personality.

9\. Legal-currentness and supersession model#

To track evolving law, systems must utilize a temporal validity schema that deprecates repealed laws (e.g., Colorado SB 24-205) in favor of their replacements (e.g., SB 26-189). A supersession model links a previous\statute\id to a current\statute\id, ensuring historical incident data remains mapped to the law active at the time of the incident, rather than retroactively applying new strict liability standards.

10\. Public wording for current legal status#

“This Machine Intelligence operates autonomously but does not possess independent legal personhood or ownership rights. Its actions are legally attributed to \[Deploying Entity Name\], acting as its principal under applicable electronic transaction laws. All assets managed by this system are held by the legal principal.”

11\. List of legally dangerous overclaims#

• Dangerous: "The protocol owns its treasury." (Correction: The legal wrapper or the token holders own the treasury; the protocol merely controls the keys).
• Dangerous: "The AI is an independent contractor." (Correction: The AI is an electronic agent deployed by a legal entity; it lacks capacity to contract).
• Dangerous: "Code is law; the smart contract limits our liability." (Correction: Code is not law; unwrapped protocols expose developers to unlimited general partnership liability).
• Dangerous: "The AI learned that behavior on its own, so we are not liable." (Correction: Under strict product liability laws like the EU PLD, post-deployment learning is considered within the manufacturer's control).

Works cited#

1. PLD — Directive (EU) 2024/2853 \- AI Resources, https://ai-resources.eu/en/atti-normativi/ue/pld/ 2. New EU Product Liability Directive Comes Into Force \- Latham & Watkins LLP, https://www.lw.com/admin/upload/SiteAttachments/New-EU-Product-Liability-Directive-Comes-Into-Force.pdf 3. xAI Sues to Enjoin Colorado's AI Act Before June 30 Effective Date, Joe Cahill \- Our Take, https://ourtake.bakerbotts.com/post/102mpre/xai-sues-to-enjoin-colorados-ai-act-before-june-30-effective-date 4. Case: xAI v. Weiser \- Civil Rights Litigation Clearinghouse, https://clearinghouse.net/case/48129/ 5. Global Legal Frameworks for Decentralized Autonomous Organizations \- Kancelaria Skarbiec, https://kancelaria-skarbiec.pl/en/global-legal-frameworks-for-decentralized-autonomous-organizations/ 6. CHAPTER 1\. General Corporation Law \- Delaware Code Online, https://delcode.delaware.gov/title8/c001/sc04/ 7. QUOINE PTE. LTD. v B2C2 LTD \- Legal Wires, https://legal-wires.com/singapore/quoine-pte-ltd-v-b2c2-ltd-5/ 8. Case Update: B2C2 Ltd v Quoine Pte Ltd \[2019\] SGHC(I) 3; Quoine v B2C2 \[2020\] SGCA(I) 2 \- SICC / Singapore Court of Appeal on cryptocurrency and mistake in contract formed using algorithm \- RONALD JJ WONG, https://www.ronaldjjwong.com/2019/04/18/case-update-b2c2-ltd-v-quoine-pte-ltd-2019-sghci-3-sicc-decides-cryptocurrency-trade-dispute/ 9. Automated Vehicles Act 2024 \- Legislation.gov.uk, https://www.legislation.gov.uk/ukpga/2024/10 10. Electronic Transactions Act 2010 — PART 2: ELECTRONIC RECORDS, | Legal Wires, https://legal-wires.com/singapore/electronic-transactions-act-2010-part-2-electronic-records/ 11. Global Electronic Signature Compliance \- SigniFlow eSignatures, https://www.signiflow.com/global-esignature-compliance/ 12. Electronic Transactions Act 2011 00-a0-01 Xml \- Western Australian Legislation, https://www.legislation.wa.gov.au/legislation/statutes.nsf/RedirectURL?OpenAgent\&query=mrdoc\_22840.htm 13. Succession & Estate Planning 2026 \- Comparisons | Global Practice, https://practiceguides.chambers.com/practice-guides/comparison/1498/18672/29271-29272-29273-29274-29275-29276-29277-29278-29279-29280-29281-29282 14. United Arab Emirates | Cryptoassets Law Guide \- The Global Legal Post, https://www.globallegalpost.com/lawoverborders/cryptoassets-law-guide-1030040461/united-arab-emirates-248284359 15. Is There a Point to the Property (Digital Assets etc) Bill? \- Oxford Law Blogs, https://blogs.law.ox.ac.uk/oblb/blog-post/2024/11/there-point-property-digital-assets-etc-bill 16. The New EU Product Liability Reform: Addressing the Digital Age | Publications, https://www.clearygottlieb.com/news-and-insights/publication-listing/the-new-eu-product-liability-reform-addressing-the-digital-age 17. EU Product Liability Directive: Responding to Software, AI and Complex Supply Chains, https://www.gibsondunn.com/eu-product-liability-directive-responding-to-software-ai-and-complex-supply-chains/ 18. What Happens If Your Product Is Defective in the EU? The New Product Liability Directive, https://getreadycompliance.eu/new-eu-product-liability-directive-defective-products/ 19. Smart Contract Legal Enforceability: When Code Isn't Law | Astraea Counsel, https://astraea.law/insights/smart-contract-legal-enforceability-code-isnt-law 20. Legal Recognition and Governance of Decentralised Autonomous Organisations (DAOs) From Code \- Maastricht University, https://www.maastrichtuniversity.nl/sites/default/files/2026-03/2026\_02%20Barbe.pdf 21. Wyoming LLC Formation for International Founders \- Sovera Global, https://soveraglobal.com/company-formation-wyoming/ 22. DAO LLC Formation Guide: Step-by-Step Wyoming DUNA Setup | Astraea Counsel, https://astraea.law/insights/dao-llc-formation-wyoming-duna-guide-2025 23. digitizing ukrainian real estate: economic potential of tokenization \- ResearchGate, https://www.researchgate.net/publication/396136782\_DIGITIZING\_UKRAINIAN\_REAL\_ESTATE\_ECONOMIC\_POTENTIAL\_OF\_TOKENIZATION 24. VT Service Provider Liechtenstein | Global Law Experts, https://globallawexperts.com/vt-service-provider-liechtenstein/ 25. TVTG Compliance Checklist for Token Service Providers in Liechtenstein (2026), https://globaladvisoryexperts.com/tvtg-compliance-checklist-liechtenstein-2026/ 26. DLT and PIL from the Perspective of Liechtenstein \- Brill, https://brill.com/downloadpdf/edcollchap-oa/book/9789004514850/BP000034.pdf 27. DAO Liability: Are Members Personally Exposed? \- Promise Legal Insights, https://blog.promise.legal/startup-central/dao-liability-personal-exposure-ooki-dao-cftc/ 28. DeFi Protocol Hacks Case Study: When “Sufficiently Decentralized” Argument Is Not Enough, https://www.buzko.legal/content-eng/defi-protocol-hacks-case-study-when-sufficiently-decentralized-argument-is-not-enough 29. Bankrupting the Matrix: DAOs and the Code \- Emory Law Scholarly Commons, https://scholarlycommons.law.emory.edu/cgi/viewcontent.cgi?article=1256\&context=ebdj 30. Bankrupt Crypto Organizations \- Carolina Law Scholarship Repository, https://scholarship.law.unc.edu/cgi/viewcontent.cgi?article=7072\&context=nclr 31. The Legal Landscape for DAOs: Key Lessons from Lido DAO and Ooki DAO | Fenwick, https://www.fenwick.com/insights/publications/the-legal-landscape-for-daos-key-lessons-from-lido-dao-and-ooki-dao 32. VARA Rulebooks 2.0 and Crypto Licensing in Dubai (AML for VASPs) | Noura Almaazmi, https://www.almaazmilawyers.com/insights/vara-rulebooks-2-crypto-licensing-aml-uae 33. Draft topics \- not final \- Wyoming Legislature, https://wyoleg.gov/InterimCommittee/2026/CompiledInterimTopicProposals.pdf 34. AI in Healthcare: The Regulatory Landscape (Federal & State) | Live Compliance, https://livecompliance.com/learn/ai-healthcare-regulations/ 35. Utah AI Policy Act Audit Evidence: Build the Complaint and Disclosure File | DeepInspect, https://www.deepinspect.ai/blog/utah-ai-ai-audit-evidence 36. Fairness / unlawful bias in the United States \- AI Laws of the World \- DLA Piper Intelligence, https://intelligence.dlapiper.com/artificial-intelligence/?t=10-fairness-or-unlawful-bias\&c=US 37. California's new automated decisionmaking technology rules: what financial institutions need to know \- Capco, https://www.capco.com/intelligence/capco-intelligence/californias-new-automated-decision-making-technology-rules 38. Utah AI Policy Act Compliance Checklist: 9 Tests for Current Disclosure Rules | DeepInspect, https://www.deepinspect.ai/blog/utah-ai-ai-compliance-checklist 39. AI Regulatory Roundup: Recent Developments in Colorado, Connecticut, and California, https://www.kelleydrye.com/viewpoints/blogs/ad-law-access/ai-regulatory-roundup-recent-developments-in-colorado-connecticut-and-california 40. Colorado Rewrites Its AI Law \- Law Offices of Snell & Wilmer, https://www.swlaw.com/publication/colorado-rewrites-its-ai-law/