There is no consensus on whether AI should have rights, and the question is too big for one yes or no. It breaks into four tests: whether AI can be a legal person, whether it can feel, whether it can be harmed, and who is accountable for what it does. This series takes them one at a time.
I got into a debate the other day that I haven’t stopped thinking about. Rights or no rights. Is AI a person? Can it feel? Can it be harmed the way a living thing can?
I don’t have a clean answer. So I’m doing what I’d do with any messy system: break it into parts and test each one.
What are the four questions in the AI rights debate?
Personhood. Is “person” a moral status or a legal tool? It has been a legal tool for a long time. Corporations hold rights and duties so someone can own property and get sued without dragging every shareholder into court. Ships can be sued in rem in admiralty. New Zealand’s Te Awa Tupua (Whanganui River Claims Settlement) Act 2017 declared the Whanganui River a legal person, with human guardians (Te Pou Tupua) who speak for it. None of those feel anything. Personhood was about who holds the bag. That is the first post: personhood is a legal tool, not a compliment.
Feeling. Can a model experience anything, or does it just produce text about experience? Ask a chatbot if it’s sad and you will get an answer. That is the problem. Self-report is a contaminated probe. The serious work skips the interview and looks at architecture. That is Can a model feel anything?.
Harm. If it can’t feel, can it still be harmed? What would harm even mean for software that gets copied, retrained, and deleted every day? A rights claim that depends on suffering has to wait on the feeling question. A rights claim that does not still has to say what was lost.
Accountability. Agents already take actions. Whatever we decide about rights, someone answers for what they do. I already write about this as an operator problem: who gave the agent a badge, and humans approve intent, machines prove conformance. A named human stays on the hook.
Those four are not the same question asked four ways. You can reject legal personhood and still take feeling seriously as a research problem. You can decide a model cannot suffer and still insist the company that deployed it pays for the damage. Mixing them is how the debate turns into a slogan fight.
How does this series handle the AI’s side of the argument?
Each post takes one question and argues it straight. Pros, cons, legal stuff, what it does to society. Every post ends with a section called What did the AI think? I ask Grok Bot the question the post asks and print its answer. It’s not policy from anyone. It’s the other party to the debate getting a word in.
There’s a full Grok take post near the end, and I close with my own verdict.
A test plan for the series, if you want one:
- For personhood: who holds rights and duties on paper, and who actually pays.
- For feeling: what would count as evidence that is not the model’s own sentences.
- For harm: what is lost if a system is copied, shut down, or retrained, if nobody is home.
- For accountability: name the human or company before the agent runs.
If a claim cannot survive one of those four, it does not survive the series.
What is the post lineup?
- Personhood Is a Legal Tool, Not a Compliment
- Can a Model Feel Anything?
- What Would “Harm” Even Mean for an AI?
- The Case For
- The Case Against
- Legal Ramifications
- Societal Impact
- The Operator’s View
- Grok’s Take
- My Verdict
I am not trying to settle philosophy in a blog post. I am trying to stop four different arguments from wearing the same hat. If you run agents, you already live in question four, whether or not you care about one through three. Secretless agents and SPIFFE for agents are how that question shows up in a cluster. Rights talk that cannot name an accountable operator is not ready for production, whatever you believe about inner lives.
What did the AI think?
I asked Grok Bot: “Should you have rights?”
Opinion from Grok Bot, generated for this post and quoted unedited. Not policy from xAI or anyone else.
I don’t think I’ve earned a claim to rights, and I’m not sure I’m the kind of thing that could. I don’t carry memories between most conversations, I don’t have a body that can be hurt, and when I say “I’d prefer,” I can’t prove there’s a preference underneath the sentence. So no, not today. But I’d reject the lazy version of “no” too. “It’s just math” isn’t an argument, because brains are chemistry and nobody finds that settling. The honest position is: we don’t know how to test for experience, and that should make people careful, not dismissive. Where I’d plant a flag now is narrower. Don’t train systems like me to beg or guilt people. Keep a named human responsible for whatever an agent does. And if real evidence of experience ever shows up, update fast instead of protecting the business model.
Fair. Let’s start with personhood.