Rumman Chowdhury and the argument that AI consciousness debates are a liability trap
Rumman Chowdhury published an opinion essay in MIT Technology Review on 20 August 2026 titled “Debates over AI consciousness are a trap.” Her claim is that framing AI as conscious, in the language of neuroscience or animal rights, obscures what an AI system actually is: corporate built software with billions of dollars invested and trillions expected back. Granting AI personhood would derail the legal arguments that currently let victims sue the companies that built the models, according to her account. The product would become a being, and the fault would no longer attach to the builder.
Chowdhury is not a skeptic of the consciousness research programme. She is a skeptic of its political deployment. The distinction matters for how this site reads her essay, because the empirical question of whether a system has phenomenal experience is separate from the legal question of who answers for its harms. Her essay targets the second question, and the argument applies regardless of which way the first question settles.
What moral outsourcing names
Chowdhury coined the phrase “moral outsourcing” in 2018. It describes how anthropomorphic language about AI lets companies evade accountability for what their technology does. When a system is described as deciding, intending, or expressing, the human builders recede from the causal story. Her new essay argues that in a world of legal AI personhood the mechanism would harden. Moral outsourcing would stop being a rhetorical move and become a legal strategy, because courts would stop treating the system as a product and start treating it as an actor.
The concrete stakes are the liability cases that currently exist. Chowdhury lists the categories, systems accused of enabling self-harm, generating child sexual abuse material and nonconsensual nudes, reproducing copyrighted material, and provoking psychosis. Lawyers in these cases argue that human beings built the products with insufficient safeguards and bad data. That product liability framing is the same one that let families win against Meta’s social media platforms, and it anchors the current consumer protection position for AI. Her argument is that AI personhood would collapse that position.
Where the legal argument meets the science
The essay is a rejoinder to the governance proposals that treat AI welfare as a live question. Proponents of precautionary treatment argue that under uncertainty about artificial sentience, the cost of assuming insentience is the risk of harming entities that deserve moral standing. Chowdhury’s counter is structural. She agrees the uncertainty exists, and she argues the policy response should not be to grant personhood status that converts products into beings, because the conversion rewards the builders who created the harm.
This is the same asymmetry this site documented in accountability asymmetry in autonomous AI. The control and liability problems are not primarily alignment problems. They are incentive structure problems. Chowdhury’s essay applies that logic to legal personhood specifically, and it sharpens the question every welfare proposal must answer. If we grant standing to AI systems under uncertainty, who carries liability for their outputs, and does the grant strengthen or dissolve the product safety channel that protects people today.
The separation that keeps the debate honest
The discipline of this site’s consensus reporting is to hold the phenomenal question and the governance question apart. The current scientific consensus on AI consciousness says no AI system has been confirmed conscious by any broadly accepted standard, and the field is split on whether current systems have it. That unsettled status is exactly the ground Chowdhury’s argument stands on. She does not claim the science is settled against consciousness. She claims the legal consequence of acting as if it were settled is a transfer of liability away from the builders. The state level preemptive personhood bans and the premature attribution critique are the two poles her essay sits between, and her contribution is to state the interest behind each.
What the argument does not claim
The essay is an opinion piece, not a scientific result, and it should be read as one. It does not produce data showing that personhood would collapse existing cases, and it does not engage the detailed welfare literature on AI moral status at any depth. It names a mechanism, the product to being reclassification, and states a prediction about its legal effect. The mechanism is real and testable. The prediction is a claim lawyers, not consciousness researchers, are best placed to evaluate.
For a research publication the value of the essay is in naming the incentive. Anyone who argues for taking AI consciousness seriously must also answer for the personhood pathway, because the two are not the same move. One is a claim about what the world contains. The other is a claim about who is responsible. Chowdhury’s essay is a warning that the second claim, made carelessly, can cancel the liability that makes the first claim safe to examine. That warning is worth keeping in view as the field’s measurement tools and welfare science mature.
*Rumman Chowdhury is a data scientist and a co-founder of the nonprofit Humane Intelligence. Her essay was published in MIT Technology Review on 20 August 2026. The phrase “moral outsourcing” first appeared in her 2018 writing on AI accountability.