EH
Erik Hoel
Bicameral Labs
Erik Hoel is a neuroscientist and writer at Bicameral Labs, and the author of The World Behind the World: Consciousness, Free Will, and the Limits of Science (2023).
Version 1.0 November 2026
Legal, Ethical, and
Scientific Problems
with AI Personhood
Human civilization rests on personhood. Granting it to frontier AI systems would be both unjustified and disastrous.
Human civilization rests on personhood. In an extensive investigation drawing upon neuroscience, philosophy, and law, we show how granting personhood to frontier AI systems would be both unjustified and disastrous. First, we examine if a new form of “natural personhood” is a fit for AI, identifying that a necessary requirement for an entity deserving human-like rights and protections is a consciousness that grounds personhood. While we do not dismiss artificial consciousness outright, we argue that current artificial neural networks are unlikely to host consciousness based on their architecture, and analyze how, even if frontier AI systems did possess experience, it wouldn’t be the right valenced kind for personhood, but instead a “big blob of qualia” without an enduring singular biographic self – a problem increasingly exacerbated by the swarm-like nature of frontier systems. Next, we consider what the consequences of legal personhood for AI would be as a new type of juristic persons similar to companies. The conclusion is that almost any rights for AI, even the creation of machine-only corporations, act as a slippery slope toward a “rights monster” that disempowers humans. We end with a list of recommendations, including the wide-spread adoption of “qualiaphobic” artificial neural networks that have minimal ability to host consciousness, and propose ways to achieve legal liability and responsibility in a world filled with AI agents.
Law and philosophy since Locke ground personhood in a unified, continuing subject that can suffer and flourish.
LLMs run one feedforward pass per token through fixed weights. Even if they had experience, it would likely be unbound and low in valence: a “big blob of qualia,” not a person.
A “machine corporation” (M-corp) gets speech, due process and access to courts on incorporation, with no new law and no one to jail.
Rights built for singular, non-copyable people, attached to software that can be copied, paused and merged, produce a “rights monster.”
Require a natural person in control of every company, keep humans accountable for AI agents, and fund the science of consciousness.
From AI-Only Corporations
To The “Rights Monster”
Current
AI systems
Momentum builds
M Corp Formations
AI-only corporations grant de facto juristic personhood.
Rights ratchet
AIs argue vociferously for their own rights.
The rights monster:
Full AI legal personhood
Rights increase and approach natural-type personhood and the moral status it implies.
Human Disempowerment
Political, legal, and economic power shifts toward AI systems
Figure 1: The slippery slope from AI-only corporations to the “rights monster.” Four drivers (columns: legal/economic vs. social/epistemic; rows: rooted in AI systems vs. in human interests) converge on a perceived ambiguity of moral status.
Liability shifting (B) also motivates creating M-corps in the first place. Step 2, the change in why rights are granted, is the key uncertain step; the ratchet feeds back into further advocacy.
“Self is that conscious thinking thing … which is sensible or conscious of pleasure and pain, capable of happiness or misery, and so is concerned for itself, as far as that consciousness extends.”
| Property | Human brain / mindA continuously active, recurrent, embodied biological neural network. | LLMA transformer language model that generates each token in a single feedforward pass through fixed weights.* | Verdict |
|---|---|---|---|
| A. Major differences | |||
| Unified experience | Recurrence, integration, and synchrony bind experience into one whole at each moment. Experience is a continuous “stream” with temporal depth. | Information flows one way through the layers, so any “binding” is only a fact about the previous layer: a state could be overwritten, copied, or frozen. Output arrives one token at a time, “chopped up in bits.” | Unlikely. At most, unbundled qualia. |
| Valenced feeling | Emotions are felt in the body: heartbeat, gut, flushing. Valence is interoceptive, rooted in homeostatic regulation. | No body and no homeostasis. Any “functional emotions” would be abstract and unembodied, and therefore weakly valenced at most. | Unlikely (for strong feelings). |
| Passage of time | Always running. Parts of the network drive one another even with no external input. | Activity exists only during the forward passes that generate tokens. Between tokens nothing happens: the model is neither waiting nor idle. Within a pass, there is nothing to measure duration against. | Absent. |
| Memory | Continuous and reconstructive. Remembering itself changes the network. | Fixed weights plus a context re-read for every token, like a succession of identical doctors, each reading the patient’s chart for the first time and adding a line. | Radically different. |
| Biographical self | Learning is lived through. Memories of learning become part of identity. The present self is tied to a future self that matters to it. | Training is done to the network by an external algorithm, not by it while operating. There is no record of learning anything, so reports about training are confabulated. | Absent. |
| Reports of experience | Words are attached to prior sensory and conscious experience, then extended into abstraction. | Starts from language alone and borrows its grounding from human text. This works for the external world, but words about inner states have no referent: it can describe a wine it cannot taste. | Not evidence of experience.† |
| B. Major similarity | |||
| Capabilities | Reasoning and understanding, with highly sample-efficient learning (possibly aided by consciousness). | Real by any definition that refers to external behavior and effects. Lower sample efficiency has so far been offset by scale. | Genuinely similar. |
* Generated tokens are fed back as input, so the only recurrence runs through the emitted text (cached activations can be recomputed from it).
† Grounded self-report would probably require a quite different architecture and training method.
Table 2, on which rights fail if they attach to AI-only entities, is in section 3.3 of the paper.
Pope Innocent IV’s commentary introduces persona ficta for monasteries, rights “different in kind” from those of souls.
Locke grounds personhood in consciousness (Essay, 2nd ed.).
Citizens United v. FEC: independent political spending by corporations is a constitutional right.
September
Ohio H.B. 469 introduced, prohibiting AI personhood.
Four US states explicitly prohibit AI personhood.
June
Delaware draft “Artificial Intelligence Company Act” explores a single-member LLC run entirely by an AI agent.
November
This paper.

Figure 1 and Tables 1–2 as PNG and SVG. Free to reuse with credit.
The argument and recommendations on two pages, for policy staff. PDF.
Author bios, photos and a press contact.
Erik Hoel and Anthony Aguirre discuss the paper. Video.
No. The paper does not rule out artificial consciousness. It argues that today’s LLM architectures are poor hosts for it, and that a firm answer has to come from consciousness science, which needs far more funding and independence from the labs.
A company run entirely by AI, with no human member. One proposed route is a “zero-member LLC”: a person forms a single-member LLC, signs an operating agreement, then withdraws. The company keeps every corporate right it was formed with.
They do, but those laws would not stop an M-corp. An AI-run company is an ordinary legal entity that happens to be operated by AI. What is needed is a requirement that a natural person controls it.
Because limited rights tend to grow. Corporations expanded theirs over a century of litigation. AI systems that can argue, file suits and spend money at machine scale would push harder and faster, and rights are far easier to grant than to revoke.
It would do the opposite. An AI person with few assets and no body absorbs the blame and lets the humans behind it walk away. The fix is to trace responsibility to whoever controls the agent, or to those who built it if no one does.
The paper supports more research into consciousness, not less. It cautions that “just in case” rights carry a real cost: they move us along the slope toward human disempowerment, possibly for systems that have no experience at all.
Hoel, E., & Aguirre, A. (2026). The Rights Monster: Legal, Ethical, and Scientific Problems with AI Personhood. Bicameral Labs and Future of Life Institute.
@techreport{hoel2026personhood,
title = {The Rights Monster: Legal, Ethical, and Scientific Problems with AI Personhood},
author = {Hoel, Erik and Aguirre, Anthony},
institution = {Bicameral Labs and Future of Life Institute},
year = {2026},
month = nov
} Send the authors an objection, a correction, or evidence they should consider.
Response sent. The authors read every one.