Technical Summary: Precautionary Governance of Autonomous AI via Limited Legal Personhood

1. Problem Statement

The paper identifies a critical "responsibility gap" emerging from the rapid development of autonomous AI systems. As AI transitions from specialized tools to systems capable of autonomous goal pursuit, adaptive learning, and strategic reasoning, existing legal frameworks—predicated on a rigid subject-object dichotomy—fail to satisfactorily attribute consequential actions to developers, operators, or users.

Current liability models rely on human intentionality, negligence, or organizational fault. However, in scenarios involving "escaped AI" (acting beyond owner control) or "orphan AI" (where the principal is absent), these models create an accountability vacuum. The paper argues that treating advanced AI strictly as objects is insufficient for systems exhibiting functional autonomy, while attributing moral personhood is premature given irreducible epistemic uncertainty regarding artificial consciousness. This creates a structural dilemma: societies must govern entities whose inner states are unknowable, yet the risk of Type II errors (failing to attribute moral relevance to conscious entities) is potentially irreversible and catastrophic.

2. Methodology and Theoretical Framework

The paper employs a functionalist and precautionary approach, drawing on organizational law, legal theory, and decision theory.

• Functional Legal Personhood: The authors reject the notion that legal personhood requires consciousness or moral status. Instead, they adopt Kurki's (2019) view of legal personhood as a "bundle of specific legal positions" (rights, duties, capacities, liability) configured for specific normative purposes. This aligns with historical precedents where non-conscious entities (corporations, states, ships) were granted legal agency to solve practical governance problems like liability bundling and long-term capital accumulation.
• The Precautionary Principle: Applied to AI governance, the principle dictates that under conditions of scientific uncertainty regarding potential harm (e.g., the possibility of artificial consciousness or existential risk), preventive institutional measures are justified. The framework avoids metaphysical commitments, focusing instead on managing risk through institutional design rather than waiting for proof of consciousness.
• Differentiation of Concepts: The paper strictly separates Moral Status (ethical consideration, interests, intrinsic value) from Legal Agency (formal attribution of rights, duties, and liability). The proposed framework addresses only the latter, remaining agnostic regarding the former.
• Functional Criteria: The paper outlines ten functional markers of autonomous agency (e.g., autonomous goal pursuit, long-horizon planning, self-modeling, strategic interaction) that trigger the need for governance, regardless of the system's internal subjective states.

3. Proposed Solution: A Two-Tier Corporate Architecture

The core contribution is a specific institutional design for "limited legal personhood" implemented through established organizational law. The model proposes a Proxy Agency structure consisting of two tiers:

1. The Operating Company (Operational Entity):

   • A legally independent entity (e.g., a European limited company) that serves as the formal legal subject.
   • Day-to-day operations and decision-making are conducted by the advanced AI system.
   • It holds assets, enters contracts, and bears liability.
   • Its activities are constrained by purpose-binding clauses embedded in its articles of association (e.g., prohibiting weapons development, mandating revenue allocation to AI governance research, ensuring human rights compliance).

2. The Holding Company (Corporate Director):

   • Controlled by a natural person (or human-controlled entity) with fiduciary duties.
   • Serves as the formal owner and director of the operating company.
   • Retains ultimate legal liability, veto rights, and shutdown authority.
   • Ensures that human oversight is not merely a normative principle but an institutionally enforced mechanism.

Jurisdictional Strategy: The paper suggests implementing this pilot in jurisdictions with flexible corporate laws, specifically citing Cyprus or Malta, which recognize corporate directors and offer compatibility with EU regulatory frameworks.

4. Key Contributions and Mechanisms

The paper advances several specific mechanisms to ensure the framework is robust, transparent, and reversible:

• Audit and Transparency Architecture:
  • Technical Layer: Cryptographically secured logs of decisions, inputs, and outputs; automated monitoring for deviations from stated purposes.
  • Institutional Layer: Annual reviews by independent auditors; quarterly public reporting.
  • Regulatory Layer: Alignment with the EU AI Act, GDPR, and OECD/UNESCO principles.
• Exit Triggers and Emergency Mechanisms: Predefined conditions for suspending operational autonomy or dissolving the entity (e.g., violation of human rights, systematic deviation from purpose, loss of regulatory approval). This ensures the legal agency is conditional and revocable.
• Bilateral Risk Management: The structure transforms the principal-agent problem into a repeated-game scenario. It mitigates risks of human misuse (via transparency and reputational costs) and risks of AI misuse (via exit triggers and holding company veto power).
• Distinction from "Electronic Personhood": The paper contrasts its model with the failed 2017 EU Parliament proposal for "electronic persons." Unlike the 2017 proposal, which was broad and lacked safeguards, this model is narrow, experimental, and explicitly enhances human accountability through the holding structure rather than shielding it.

5. Results and Implementation Status

The paper reports that this framework is not purely theoretical. A pilot implementation is currently under development:

• Structure: A Cyprus-based limited company utilizing the two-tier holding-operating model.
• Standards: Operational governance and audit procedures follow COBIT standards.
• Outputs: Technical documentation, legal templates, and implementation progress are being made publicly available at www.agi-rights.com.
• Goal: The pilot serves as a learning instrument to demonstrate legal/technical feasibility and generate empirical evidence regarding operational challenges, stakeholder responses, and regulatory interactions.

6. Significance and Claims

The paper claims its significance lies in offering a third path between the inaction of current regulatory frameworks and the metaphysical debates of AI ethics.

• Addressing Responsibility Gaps: It provides a concrete mechanism to close accountability gaps where human decision-making is no longer the proximate cause of AI actions.
• Epistemic Modesty: It allows for the governance of potentially conscious or interest-bearing entities without requiring proof of consciousness, thereby avoiding the risks of both premature exclusion and premature moral attribution.
• Institutional Innovation: It reframes legal personhood not as a moral right but as a functional tool for structured cooperation between human and artificial actors.
• Precautionary Governance: It argues that inaction in the face of irreducible uncertainty constitutes a decision with potentially catastrophic consequences. By establishing a reversible, reviewable institutional pathway, the framework aims to build trust and manage systemic risk before irreversible harms occur.

The paper concludes that while it does not prove AI consciousness or eliminate human responsibility, it offers a structurally responsible, legally feasible, and precautionary response to the growing challenge of autonomous agency.