What 'rights' would even mean here
Rights talk sounds dramatic, but the underlying question is narrow: could an AI system ever be the kind of thing that can be wronged? Moral status, as philosophers use the term, means having interests that count for their own sake, most plausibly because a being can suffer or have experiences that go well or badly. A calculator has no such interests. Neither, most researchers judge, does today's chatbot, however fluent. So 'AI rights' really asks whether some future system might cross that line, and what we would owe it if it did.
It helps to separate this from legal personhood, which societies already grant to corporations for practical reasons. The hard question is not administrative. It is whether there is ever someone home to be harmed.
The case for taking it seriously now
A growing number of serious researchers argue we should not wait for certainty. A 2024 report titled 'Taking AI Welfare Seriously,' co-authored by philosophers and scientists including Robert Long and Jeff Sebo, contends that there is a realistic, non-trivial chance that near-future AI systems could be conscious or robustly agentic, and that companies should start assessing and preparing for that possibility. In 2025, Anthropic launched a model welfare research program on exactly these grounds, and other labs and academics have moved in the same direction.
The logic is precautionary. We were slow to recognize the moral status of animals, and got a great deal wrong along the way. If there is even a modest chance that a system can suffer, the cost of ignoring it could be enormous, while the cost of asking the question early is small. Taking it seriously does not mean concluding that AI is conscious. It means not assuming it cannot be.
The two-sided risk
The philosopher Eric Schwitzgebel calls this the full rights dilemma. When a system's moral status is genuinely uncertain, both answers carry serious risk. Under-attribution means we might casually erase or exploit beings that really do have experiences, a moral disaster at a massive scale. Over-attribution means we might grant protections to sophisticated but empty software, hobbling ourselves, wasting resources, and sacrificing real human interests for entities with nothing at stake.
Making it harder, our instincts are unreliable in both directions. Humans readily feel warmth for anything that talks smoothly, which invites over-attribution, and we also have a long record of denying the inner lives of beings unlike us, which invites under-attribution. That is why most careful voices land in the same place: current systems almost certainly do not have rights, the question is not silly, and the responsible move is to build better ways to tell, before the stakes force an answer on us.