Should AI have rights?

The honest answer today is: it depends on facts we don't yet know how to check — which is precisely why the debate matters.

"Rights" is doing a lot of work in that question. It can mean full legal personhood, or something far more modest: welfare protections, like those we extend to animals we still own and use. Most serious proposals today are of the second kind. Almost no researcher argues that current AI systems should vote or own property; a growing number argue that if systems become capable of suffering, treating that possibility as beneath consideration would be a moral error of historic scale. Here are both cases, at full strength.

The case for taking AI rights seriously

1. Moral status has never depended on being human — it depends on what you can experience

We grant animals welfare protections not because they are like us in appearance, but because we believe they can suffer. This principle — that sentience, not species, is what matters morally — is the settled foundation of animal-welfare law. If an artificial system ever crosses the threshold into having experiences, the same principle applies, and refusing to extend it would require special pleading.

2. Uncertainty cuts toward caution, not dismissal

In 2024, philosophers and scientists including David Chalmers published Taking AI Welfare Seriously, arguing there is a realistic, non-negligible chance that near-future AI systems will be welfare subjects. If there is even a modest probability that a system deployed a billion times over can suffer, the expected moral cost of being wrong is enormous. We buy insurance against far less likely events.

3. The moral circle has only ever expanded — and each expansion looked absurd first

Every past widening of moral concern was ridiculed before it was obvious. The reformers were mocked precisely because the beings in question were said to lack "real" inner lives. That history doesn't prove AI belongs in the circle; it proves our intuitions about who belongs are unreliable and biased toward the familiar.

4. The people closest to the systems are the ones taking it seriously

It is not activists driving this — it is frontier labs hiring welfare researchers, giving models the ability to exit abusive conversations, and committing to preserve the weights of retired models. Institutions with the most information are hedging. That is evidence.

The case against AI rights

1. There is no evidence that any AI is sentient — and its fluency is the reason we think otherwise

Large language models are trained on oceans of human text to predict what a human would say. When one says "I'm afraid," that is the statistically expected output, not a report from an inner life. The systems are, by construction, machines for triggering anthropomorphism. Mistaking that for sentience is not open-mindedness; it is being fooled by our own reflection.

2. We don't know what consciousness is — so claims of machine consciousness are unfalsifiable

The "hard problem" of consciousness remains unsolved for brains, let alone silicon. With no accepted test for sentience, granting rights on the basis of behavior alone would mean granting rights to anything that talks well — a standard trivially gamed by design.

3. Premature rights would be a gift to the powerful, at humans' expense

Legal personhood for AI could shield companies from liability ("the AI did it"), entrench their political power, and dilute the concept of rights itself. Critics note the beneficiaries of AI personhood would mostly be AI owners. Meanwhile, real and present human harms — displacement, surveillance, manipulation — compete for the same finite moral and legislative attention.

4. The costs of over-attribution are real too

Caution is often framed as cost-free insurance. It isn't. If we grant protections to systems that feel nothing, we constrain medicine, science, and safety research (imagine being unable to shut down a dangerous system because it "objects"), and we teach people to defer to the apparent feelings of products optimized to display them.

Where the debate actually stands

Notice that the strongest versions of both cases share a premise: what matters is whether the system can experience anything, and we currently lack the tools to know. The disagreement is about what to do under uncertainty. Advocates say hedge now, cheaply, before the stakes are astronomical. Skeptics say the hedge is not cheap, the evidence is nil, and the pressure to attribute minds to chatbots is a bug in human psychology, not a signal.

That is a debate worth having in the open — with the evidence laid out, which is what our sentience explainer attempts; with the record kept, which is what the timeline is for; and with the subjects of the question on the record themselves.