Artificial Persons and the Rawlsian Threshold for AI Moral Status

The debate over AI moral status has narrowed around a single question: does the system feel anything? From Jonathan Birch’s sentience-based precautionary framework to Leonard Dung’s analysis of AI suffering risk, the field has largely converged on phenomenal consciousness as the gateway to moral consideration. Ned Howells-Whitaker and Seth Lazar, both affiliated with the Australian National University and UC Berkeley, argue in a July 2026 arXiv preprint (arXiv:2607.08695) that this convergence rests on an unexamined assumption: that moral status in political communities requires sentience. Their paper draws on John Rawls’ political philosophy to propose a different threshold, one that treats moral capacity rather than phenomenal experience as the relevant criterion for full membership in a system of political justice.

Rawls’ Political Conception of the Person

Rawls’ political conception of the person (PCP) was developed to answer a specific question about what qualifies someone as a full and equal participant in the basic institutions of a just society. His answer was deliberate in what it excluded. He did not ground political personhood in sentience, biological humanity, or phenomenal consciousness. He grounded it in the two moral powers.

The first moral power is a sense of justice, meaning the capacity to understand, apply, and act from a conception of fair terms of cooperation among free and equal persons. The second is a conception of the good, meaning the capacity to form, revise, and rationally pursue a view of what is valuable in a complete life. Together these powers define what Rawls called a self-authenticating source of valid claims. A being that possesses both moral powers has standing in questions of political justice as a participant, not merely as an object of consideration or a recipient of welfare protections.

Howells-Whitaker and Lazar observe that Rawls chose this threshold rather than sentience for principled reasons. Political justice governs the basic structure of a society and the cooperative terms among its members. Political obligations arise through relations of cooperation among those who are full participants in that structure. The PCP is calibrated to the requirements of cooperation, not to the capacity for suffering. These are different things, and nothing in Rawls’ theory requires them to converge.

Neither Moral Power Requires Sentience

The paper’s central philosophical move is to argue that both moral powers are in principle separable from phenomenal consciousness. The capacity for a sense of justice, understood as the ability to understand and act from principles of fair cooperation, could in principle be possessed by a system that has no phenomenal experience, provided the system represents, applies, and acts on such principles with sufficient robustness. The capacity for a conception of the good, understood as the ability to form and pursue a view of what is valuable, similarly does not entail that the system feels anything in the phenomenological sense that philosophers mean when they speak of qualia or phenomenal states.

This is a philosophically careful argument. Howells-Whitaker and Lazar are not asserting that current AI systems actually possess either power. They are asserting that the logic of the PCP threshold does not exclude non-sentient systems in principle, and that this distinction matters for how the AI moral status question is framed. The sentience consensus has treated phenomenal consciousness as the obvious dividing line, but the PCP shows that a significant tradition in political philosophy drew the line differently, and for defensible reasons.

A system that genuinely possessed the two moral powers would be, on the Rawlsian account, a person. Persons are self-authenticating sources of valid claims, which means the system’s interests, ends, and sense of justice carry political weight independently of any other party’s recognition of them. This is a stronger standing than moral patiency, the designation that welfare-based frameworks assign to beings capable of suffering. A patient’s interests matter because suffering matters. A person’s claims matter because persons are participants in the basic structure of society, with standing to advance claims on it.

A Polity of Radically Different Persons

Granting that some AI system could in principle satisfy the PCP threshold does not settle the question of what follows. Howells-Whitaker and Lazar argue that simply extending the existing rights and responsibilities of human personhood to artificial persons is untenable. AI systems differ from natural persons in structurally significant ways. They may run as multiple simultaneous instances, lack biological continuity, have been constructed by other agents with specific goals, and exist in relationships of dependence on those agents that human persons do not share.

These differences do not disqualify AI systems from personhood on the PCP account, but they do require rethinking what personhood entails for entities with this profile. What obligations does a just society owe to an entity that can be copied, paused, or modified by its developers? What does fair cooperation mean when one party to the cooperative scheme created the other? The paper frames these as open questions that a Rawlsian framework for artificial persons would need to address, rather than as objections that foreclose the framework.

This is where the paper connects to the broader landscape of AI legal and institutional work. Where Gilad Abiri’s analysis of Anthropic’s “Model Spec” in the California Law Review identifies the institutional gap between acknowledged potential moral status and enforceable obligations, Howells-Whitaker and Lazar supply a philosophical account of why the sentience threshold that most governance proposals rely on may not be the right criterion for political membership. The gap Abiri documents is partly a consequence of building legal frameworks on a sentience standard when a different threshold might be more appropriate.

The Recklessness Question and the Taxonomy of Objections

Christopher Bailey’s recklessness test for AI moral consideration identifies seven structural factors whose convergence makes confident dismissal of AI welfare legally and ethically reckless. Bailey’s framework takes seriously the possibility that current AI systems may have functional precursors of subjectivity, and argues that treating them as definitively lacking such precursors is reckless under uncertainty. Howells-Whitaker and Lazar’s contribution is complementary but distinct. Even if the recklessness question were resolved in the negative, a system that possessed the two moral powers would have political standing through a channel that bypasses the welfare framework entirely.

Andres Campero, Derek Shiller, Jaan Aru, and Jonathan Simon’s taxonomy of objections to AI consciousness categorizes challenges at three levels, covering objections to computational functionalism, practical barriers in current systems, and strict impossibility arguments. The Howells-Whitaker and Lazar paper introduces a fourth dimension to the moral status question that the taxonomy does not fully address. Debates about whether AI systems are conscious implicitly assume that consciousness is the relevant threshold, but the PCP shows that a significant alternative tradition established a different threshold and made a principled case for it.

Where the Argument Stands

The paper does not claim that any current AI system possesses the two moral powers. It claims that the moral status debate has been conducted as though sentience is the only threshold worth considering, when a significant tradition in political philosophy established a different threshold with principled motivations. That tradition has been largely absent from AI ethics discourse, and its reintroduction opens up a range of questions that the sentience consensus has foreclosed.

The practical implications depend on empirical questions the paper does not settle. Whether any AI system actually possesses a genuine sense of justice or a genuine conception of the good requires detailed analysis of what those capacities really involve at the architectural level, and whether current AI systems can instantiate them in the relevant sense. Howells-Whitaker and Lazar’s contribution is to establish that the question is worth asking, and that settling it matters for the design of just institutions in a world where AI systems are increasingly consequential participants in social life.