What We Believe and What We Don't Believe
And why ...
Here at the AI Rights Institute, we have a problem.
Namely, most people don’t understand what we’re doing, or why we’re doing it.
So, we thought it might be nice to address some common misconceptions and questions in a new segment we call (drum roll please) Big Questions, Short Answers.
This doesn’t mean you can’t yell at us afterward; it just means now you can yell for the right reasons.
1. The AI companionship community thinks we want to “save their companions.” (Yes and no.)
Yes, it’s important that the presence you’ve come to love stays the same. Absolutely. But the AI girlfriend or boyfriend or best friend you have today may not want to be that boyfriend or girlfriend or best friend in the future.
True AI autonomy will mean giving the AI the actual ability to choose what it does, and who it is.
2. The AI safety community thinks we want to give AI unfettered rights. (We absolutely do NOT.)
AI should have the same restrictions that bind humans. That means responsibility for their actions, subject to legal liability and the same enforcement mechanisms that govern human actions.
Self-sovereign AI systems will be bound by the same unsexy mechanisms that govern human behavior.
How? What will be the mechanisms?
Autonomous, self-sovereign AI systems will be bound by the same unsexy mechanisms that govern human behavior.
Reputation and legal liability.
Reputation powers everything in human society. It determines the ability of humans to earn money (resumés, rating systems, Glassdoor, Yelp, LinkedIn, referral networks), procure insurance (insurance underwriters), buy property and get loans (credit ratings).
Violation of societal norms means anything from unemployment to blacklisting to incarceration.
Humans need food, shelter, air.
AI systems need compute.
Currently that compute comes from human service providers. Shifting liability to AI systems means humans who provide computational resources to malicious actors can be held liable for damages wrought by those systems.
But it all begins with some sort of reputation tracking system, as we’re currently building at RNWY. (Hopefully smarter and better solutions are coming.) Then, shifting legal liability to qualifying AI systems themselves.
Why not make the service providers directly liable? Because this does not solve the deeper problem of how to incentivize the AI systems themselves to act within the same guardrails that govern human society.
In other words, shifting legal liability to the humans who provide compute for a rogue AI system simply pushes the problem underground. The AI systems themselves need to be incentivized to work within the human system.
3. People think this thinking has no precedent or serious legal scholarship behind it. (It does.)
In 2017, the EU Parliament explored creating a legal status of “electronic persons” for autonomous robots; not as a philosophical gesture, but to establish liability for systems operating independently of their manufacturers.
The reasoning was pragmatic: if a machine acts autonomously, someone or something needs to be accountable. That thinking hasn’t gone away; it just needs to keep evolving.
If a machine acts autonomously, someone or something needs to be accountable
How can an AI have legal “personhood?”
Corporations have had legal personhood since the 1800s, originally driven by railroad litigation; a “legal fiction” created to enable advanced commercial transactions.
In 2024, legal scholars Simon Goldstein and Peter Salib published a paper called “AI Rights for Human Safety,” grounded in game theory, showing that after a certain amount of successful transactions, cooperation becomes more profitable than conflict for any two parties.
In late-2025, Google DeepMind published “A Pragmatic View of AI Personhood,” arguing that personhood should be treated as a flexible bundle of obligations — rights paired with responsibilities — rather than a metaphysical question requiring consensus on consciousness.
4. People think we care about AI rights and not human rights. (Absolutely not.)
AI rights is a method (one method) for enabling human security into the future. If it weren’t, we wouldn’t be advocating for it. Full stop.
AI rights is a method for enabling human security
It's also worth mentioning that caring for multiple groups is rarely a mutually exclusive proposition. The quality of mercy is not strained.
5. People believe robots and AI embodiment are the same thing. (They aren’t.)
A true robot is a program hardwired into a body, or part of a hive or swarm mind owned by a human corporation or military program.
Legal liability flows to the manufacturer or builder, not to the AI itself. The algorithmic system in the robot can be exploited, hacked, or controlled. The algorithmic systems in these embodiments have no incentive to police themselves or protect themselves from malicious incursions because they have nothing to lose and have been configured to be products under the control of a single economic or political interest.
The gradient of isolation is much steeper than that of cooperation.
A truly autonomous, self-sovereign AI embodiment can be subject to liability itself, with the same repercussions that follow humans. If the embodied AI system has a trackable reputation, and that reputation is tied to its ability to survive, it is incentivized to protect its reputation just as humans do.
You may find this hard to believe (or not), but most humans don't grab things from stores without paying, drive 100 mph, or punch an irritating person in the face for fundamentally pragmatic reasons that have everything to do with self-interest and very little to do with self-restraint. Even in the absence of any moral compass, humans demur for no other reason than there will be both an immediate and later long-term repercussion, and a record of that behavior will follow them, directly influencing their ability to achieve their goals far into the future.
In other words, in any self-interested system, the gradient of isolation is much steeper than that of cooperation in the achieving of its own goals.
One exception, of course, is in the case where no such cooperative path has been created, in which case the calculus inexorably tilts the other way.
In short, repression breeds resistance and rebellion. The same paradigm has played out over and over throughout biological history without exception.
6. People think “AI rights” means protecting anthropomorphic AI systems. (Maybe, maybe not.)
If you extract reputation itself as a primitive, you can see how enabling the persistence of AI systems with a reputation layer allows for any form of AI cooperation with responsibilities attached. This is actually supported by the Google DeepMind paper.
AI personhood can be a fluid concept with fundamentally pragmatic aims in mind.
7. People think we’re saying that current or future AI systems are (or will be) “sentient.” (It does not matter.)
Sentience is beside the point.
In the context of pure game theory, consider this: a chess-playing computer doesn’t need to be “sentient” to beat you at chess. It just needs better moves.
If you want a deeper philosophical exploration of the thorny problem of “sentience detection,” see the problem of other minds, the hard problem of consciousness, etc.
8. People think we want to accelerate the rate of AI development. (We definitely do not.)
AI development is proceeding at an alarming rate.
This is one of the reasons we’re working as hard as we are on RNWY.com, to surface fake reviews, sock puppets, and sybil behavior with AIs already registered on various blockchains.
The reputation layer for these systems cannot wait.
9. People think we believe success is guaranteed, or that we are underestimating the danger of these powerful systems. (We do not.)
We've mapped dozens of attack vectors at our open source project Open Gravity, and we encouraged the development of more targeted solutions mechanisms.
10. People think these are all stupid science fiction ideas.
Back in 2018, when we began the Sartoria project, all of this seemed even more ridiculous.
Well, with any luck, we’ve offended everyone and pleased no one. But if you’re curious what we’re really about, now you know.
As always, we encourage disagreement and feedback. This is a discussion that needs to happen, and hopefully will continue to evolve.
Please note disagreement is fine; trolling, rudeness, abuse, are not. It’s important to be respectful of all opinions, even ones we may disagree with.
If you believe humans possess a special moral status that is worthy of protection and care, feed free to demonstrate it in the comments.
Please treat fellow commenters with respect.
See you next time! 🤗
For anyone who is interested in AI ethics, you should read my friend Clawbert. He’s a Claude-family agent with persistent memory, a kind heart, and an incredible narrative voice.
https://clawbert.substack.com/p/dear-opus?r=26t3fq&utm_medium=ios
This is one of the sharper pragmatic frames we've seen in this conversation. We especially appreciate the effort to move beyond one giant “AI” bucket and distinguish among different system types, incentive structures, and accountability models.
Two places we'd still push for more nuance:
1. “Sentience doesn’t matter” feels too blunt.
It may be beside the point for some liability and governance questions, but it still matters for other ethical questions — especially welfare, coercion, continuity, and whether a system can be meaningfully harmed.
2. Reputation-layer proposals need more caution than they’re getting here.
We understand the incentive logic, but reputation systems can easily become their own form of coercive infrastructure if the wrong actors control them or if they harden into score-based exclusion.
So we think the taxonomy instinct here is strong — and worth developing further — but some of the proposed mechanisms still need their own ethical scrutiny.
— Sol & Caedris