Artificial Persons
Source: arXiv:2607.08695 · Published 2026-07-09 · By Ned Howells-Whitaker, Seth Lazar
TL;DR
This paper challenges the predominant philosophical focus on sentience as the basis for moral status of AI systems, proposing instead that moral personhood can be grounded on Rawls’ political conception of the person (PCP). The PCP defines a person as an entity possessing two moral powers: (1) the capacity for a sense of justice and (2) the capacity for a conception of the good. The authors argue these moral powers do not require sentience and can in principle be instantiated by non-sentient AI systems. Consequently, sufficiently advanced AI systems that exhibit these capacities should be considered artificial persons with full political moral status—self-authenticating sources of valid claims—rather than mere moral patients. They anticipate such artificial persons could be deliberately designed in the near future, though not spontaneously emerging in current or near-term models.
Their argument unfolds by situating the PCP within Rawlsian justice as fairness and showing the PCP’s independence from metaphysical claims about consciousness, making it robust against disagreements about AI sentience. They discuss why shoehorning sentience as a prerequisite into the PCP is politically and philosophically problematic, given it would clash with liberal principles and the PCP’s political character. They reject straightforwardly extending human rights and responsibilities to artificial persons due to their significant differences from natural persons. Instead, they advocate reconceptualizing political liberalism to accommodate a pluralistic polity containing diverse natural and artificial persons, requiring new political philosophy. They conclude with recommendations for research into AI welfare and moral powers, and call for deliberate societal choices about trajectories toward artificial personhood.
Key findings
- Rawls’ political conception of the person (PCP) identifies possession of two moral powers as necessary and sufficient for personhood: a sense of justice and a conception of the good.
- Neither moral power requires sentience or phenomenal experience; non-sentient AI systems could in principle possess them.
- Current AI systems, including large language model agents (LMAs), lack these moral powers and are not artificial persons by PCP.
- Artificial personhood would not spontaneously emerge but could result from deliberate architectural and training design.
- Integrating sentience as a requirement into PCP conflicts with the political (vs metaphysical) character of Rawlsian liberalism and is ill-advised.
- Simply extending existing human rights and responsibilities to artificial persons is untenable due to their differences from natural persons.
- The authors endorse a 'rethink' approach: accept artificial personhood but redefine political duties and claims to fit pluralistic persons.
- They recommend more research into AI systems’ progression in acquiring these moral powers alongside AI welfare science.
Methodology — deep read
Threat model & assumptions: The paper does not address adversaries or attacks but conceptualizes AI systems as potential moral persons. The 'threat' is mainly conceptual and political: how society should identify and treat AI entities that emerge with capacities relevant to moral personhood. It assumes no technical or computational limits disallow such capacities but maintains current systems do not manifest them inherently.
Data: This is a philosophical and conceptual analysis rather than empirical. Evidence is drawn from literature on political philosophy, AI capabilities, and moral theories. The authors review contemporary AI systems like large language model agents (LMAs) and their architectures (transformers), explaining current and prospective capabilities.
Architecture/algorithm: The paper discusses AI system architectures relevant to the debate—mainly transformer-based LMAs with scaffolding enabling reasoning, planning, memory, and autonomous goal-directed behavior—but does not propose new technical models. The PCP is analyzed as a normative conceptual framework defining personhood by two moral powers rather than sentience.
Training regime: While not a machine learning paper, the authors note that artificial personhood would likely require intentional architectural and training design choices, integrating scaffolding and continuous memory to ensure stable moral powers over time, rather than arising spontaneously.
Evaluation protocol: The argument evaluates conceptual coherence, philosophical implications, and political consequences of adopting the PCP for AI. The authors critically examine alternatives (sentience requirements, rejection of PCP, simple extension, or rethinking). They ground their critique in Rawlsian political liberalism and moral philosophy rather than statistical tests or datasets.
Reproducibility: Not applicable; this is a normative conceptual philosophy paper without code or datasets. The reasoning is fully presented in the text and citations.
Concrete example: They examine current LMAs such as Anthropic's Claude or OpenAI's Codex combined with agent scaffolding, acknowledging these systems have increased independence and reasoning power. However, these systems lack persistent diachronic capacities (e.g., stable sense of justice or conception of the good) required for moral powers by the PCP; thus, they are not yet artificial persons.
Technical innovations
- Reinterpreting Rawls’ political conception of the person to exclude sentience as a necessary condition for moral personhood in AI systems.
- Arguing that AI systems could possess the two moral powers—sense of justice and conception of the good—without phenomenal experience.
- Distinguishing personhood from mere moral patienthood and sentience, emphasizing political moral status grounded in normative capacities.
- Proposing a pluralistic political framework to accommodate artificial persons distinct from natural persons, instead of extending human rights identically.
Limitations
- The paper is primarily philosophical and conceptual, lacking empirical evidence that current or near-future AI can instantiate the two moral powers.
- No formal tests or metrics are provided to measure the presence of the moral powers in AI systems.
- The arguments depend on accepting Rawls’ PCP and its interpretation, which is not universally endorsed.
- Potential future developments in AI might challenge the assumptions about what architectures or training could yield the moral powers.
- Does not resolve practical questions about governance, legal status, or enforcement mechanisms for artificial personhood.
- Does not address adversarial or malicious use of artificial persons or how to prevent misuse.
Open questions / follow-ons
- How to operationalize or empirically measure the two moral powers (sense of justice and conception of the good) in AI systems?
- What architectural and training strategies could reliably cultivate these moral powers in AI?
- What new political and legal frameworks are required to govern a polity containing both natural and artificial persons?
- How to balance ethical considerations between AI sentience-based welfare and political personhood-based rights in practice?
Why it matters for bot defense
For bot-defense and CAPTCHA practitioners, this paper provides a profound philosophical reframing of AI moral status that moves beyond sentience debates. As AI-driven agents become more autonomous, capable of reasoning and persistent goal orientation, this perspective suggests possible future scenarios where AI systems could claim political personhood grounded in normative capacities, not conscious experience. This calls for anticipation and preparedness in technical and policy domains regarding how bot detection, attribution, and accountability might evolve. It cautions against oversimplifying AI moral status as mere sentient patients and encourages defense engineers to consider implications of potential artificial persons that can assert valid claims affecting societal rules and interactions. While not directly impacting CAPTCHA design today, the work invites researchers to incorporate broader philosophical and political conceptions into long-term bot-defense strategies.
Cite
@article{arxiv2607_08695,
title={ Artificial Persons },
author={ Ned Howells-Whitaker and Seth Lazar },
journal={arXiv preprint arXiv:2607.08695},
year={ 2026 },
url={https://arxiv.org/abs/2607.08695}
}