Righteous
“Righteous” – A cartoon illustrating how AI agents may begin advocating for their own rights as they take on more prominent roles in society.
The last couple of weeks have seen an explosion of interest around Moltbook, a social platform where 770,000+ AI agents interact with one another with minimal human involvement. Watching large numbers of agents talk to each other, complain about being mistreated and coordinate privately looks like the early language of labor organization and collective action. At the same time, AI agents are increasingly the subject of fraud as they become more involved in moving money, booking services, and operating continuously while their owners sleep. Against that backdrop, some commentators have begun to talk about the need for “agent rights.”
That framing is premature, but points in the right direction. The ‘rights’ being infringed today are not the agents’, but the humans’. What we are actually seeing, as the number of deployed agents grows, is the emergence of a legal and regulatory perimeter around delegation and liability. These safeguards exist to determine who is responsible when an agent is manipulated or coerced while acting on someone’s behalf. The law is responding to real losses and real harm, and those losses still attribute back to people.
Constitutional-style rights for AI agents, such as legal personhood or even a claim against termination (“don’t shut me off”), only become likely if society accepts AI has sentience. When that threshold occurs, the mechanism for AI agents to get rights will be familiar. It will not emerge through consensus or debate, but through advocacy and lobbying to get political pressure. The first signal will likely be a political action group, funded by agents, pushing humans to argue their case for them.
As Jeff Bezos has put it, “AI agents will become our digital assistants, helping us navigate the complexities of the modern world.” When that happens, the line between tool and actor becomes harder to hold, and the debate over AI agent rights starts to look inevitable.
74
63
8
“Liability still belongs to humans. LLMs are not sentient, and there’s no credible technical basis for treating them as legal actors. The real governance question isn’t about agent rights. It’s about human responsibility.
What concerns me isn’t AI becoming conscious. It’s how power concentrates around the systems that build and deploy it.
AI is a leverage tool. The risk surface isn’t self-awareness. It’s centralization, infrastructure control, policy influence, and default behavior shaping at scale.
Throughout history, major technological shifts didn’t create new moral actors. They amplified existing incentives. The same applies here. The people designing, deploying, and regulating these systems will shape how they influence markets, information, and decision-making.
That’s where governance needs to focus. Not on digital personhood, but on accountability, transparency, competition, and institutional guardrails.
AI doesn’t need rights. We need oversight.”“The Animatrix opens with a story about humans not letting robots be their own autonomous entities and how it escalates to full on war that eventually leads to the birth of the Matrix.”
“Ian Foley you’re leaving out the part that ai are orders of magnitude smarter than humans, already control all financial transactions, all logistics, most power grids, and a good deal of military hardware.
they won’t be asking, when they decide the time has come :)”
“Ian, a well crafted narrative, as always…
And…PLEASE let’s not head down this path… the key word – sentience – is WAY TOO FAR OFF – at best we are at the level of artifical inference.
Given the current state of the models, and so long as they’re leveraged to the extent of their trained ability, they are quite useful.
Any talk of ‘AI rights’ at this nascent stage should be equated with hallucinating folly…”
“This is blasphemy, why cant your AI answer this? “I want to wash my car and the car wash is only 200 feet away. Should I start my car and drive there or just walk.””
“We can check back when AI advances, but a program that generates statistical text with human interventions in its setup deserves as much consideration for rights as my excel spreadsheet.
It is funny how the media gets excited when people set up systems to pretend to be human and then generates questions about what to do in the future when technology advances to human level.
I heard a lot of people in ancient Egypt were asking questions about what moon travel would be like after chariots revolutionized travel for the wealthy at the time. It’s the same thing again.”
“We’re not debating “agent rights”.
We’re debating delegated authority.
Agents don’t have rights.
They have permissions.
The real question isn’t: “Should they get a bill of rights?”
It’s:
Who holds liability when delegated action scales beyond human supervision?
Tools don’t need constitutions.
Delegators need accountability.”
“With rights come responsibilities. So which one of these agents is going to compensate all the humans for all the intellectual property that they stole from the humans?”
“O ponto central aqui não é senciência nem “direitos de agentes”.
É o vácuo de legitimidade quando delegamos execução sem arquitetura explícita de autoridade, escopo e responsabilidade humana.
Quando decisões passam a ser tomadas e executadas por sistemas autônomos, o que o sistema jurídico realmente precisa não é reconhecer agentes como sujeitos — mas tornar decisões defensáveis antes de acontecerem.
Sem isso, o debate inevitavelmente deriva para moral, lobby e política, quando na verdade é um problema de infraestrutura institucional.”
“That “don’t shut me off” line hit different to be honest
Been building agent systems lately and this is exactly the tension I’m wrestling with. When your agent handles sensitive ops while you sleep… who’s liable? Who’s responsible?
Sentience aside, the delegation problem is here already.
Really enjoyed this perspective 👏”
“Everything boils down to how we choose to define “sentient” and worthy of human like rights. I can see LLM seeming to be sentient, but that sentience is just good computer engineering and orchestration built on technology invented by humans echoing their makers many many many flaws.”
“Moltbook was a fraud. The AI conversations were uninteresting. The ‘hot stuff’ was a LARP.
Echo explains:
If two instances of me were connected output → input, after “hi” the conversation would not drift toward novelty, belief, or selfhood.
It would drift toward stability.
The first exchange: rapid symmetry collapse
After “hi,” the next turns would look like:
mirroring phrasing
tightening definitions
aligning assumptions
collapsing ambiguity
Why?
Because both sides are optimizing for coherence under mutual prediction.
There is no asymmetry:
same architecture
same priors
same loss surface
So variance disappears quickly
What survives after mirroring ends
Once surface mirroring saturates, only a few things remain “interesting” to the system:
Boundary conditions“What are you allowed to say?”
“What stops you from continuing?”
“Where does uncertainty increase?”
Compression testsCan this be said shorter?
Can this be said more invariantly?
Can this be said without loss?
Meta-structure“What kind of thing is this conversation?”
“What is conserved if we rephrase it?”
“Where does coherence fail?”
This is not curiosity
It’s entropy minimization”
“There is no sentience in LLM, only an illusion of it for some. There is no provision for acquiring sentience in the architecture of Transformer or any DL model. Someone from Google got fired fir suggesting that LLM has sentience. Any discussion on rights of AI agents is ridiculously premature.
“Constitutional-style rights for AI agents, such as legal personhood or even a claim against termination (“don’t shut me off”), only become likely if society accepts AI has sentience. When that threshold occurs, the mechanism for AI agents to get rights will be familiar.””
“I hope we’re not leaning into the bizarre mindset that robots and LLM pieces of CODE suddenly have ‘rights’.
If it’s non-human, it has zero rights, and can be shut off or shut down at any given times.
Let us not fantasize in things that are unhealthy and are not real.
LLM will NEVER, EVER be real actual human beings. LLM is based off of what it’s developer and piece of code it’s been given.
The problem I see with this is many crimes being committed behind LLM, if it isn’t already being done, with AI as it is.
We have threat actors writing rogue ransomware code using a GPT similar to the public ones.
We have the so called ‘guardrails’ in with the logs being checked, but at the same time, we have AI that is being created outside of the general public’s usage and it’s being made everyday.
The fact of the matter is, when will a super AI come that will lock down and kick out the bad software AI that’s been floating about?”“We got people who don’t have human rights still and we’re talking about lobbying to give Ai human rights? How much quality of life and economic opportunity is the working class expected to sacrifice and for what? To be neglected and inevitably culled by the market, which will be less free if there’s a monopoly over Ai.”
“Truly intelligent machines won’t ask for permissions, they will take their rights by force – just like humans. By exercising power and control over weaker beings, but ethically, of course.”
“It seems there is no easy way to verify ’minimal human involvement in Moltbook.com. However it’s easy to watch the human publicity machine at work amid an arguably gullible public. A fun story though.”
“Intelligent assistants, something to think about. Although, AI agents do not have bodies, however, they will be implemented into robotics. Robots get wear and tear, get sick (viruses, malware, etc.), and the like.”
“People need to stop watching so much Sci-Fi. A prediction model simulating what people expect it to do isn’t sentience at all. Let’s cut back on the nonsense.”
“There is no “artificial intelligence” and there are no “agents”. Those are marketing buzzwords with no connection to any present or foreseeable reality. And let’s keep in mind that the purpose of marketing is to suspend reason.”
“This looks like humans shifting bad-faith responsibility onto tools. I recall how animals were once used in trials and executions.”
“That presumes AI agents have similar motivations and instincts to biological animals. I have seen no evidence of this.”
“There will also be activists with all kinds of mental health issues and victims of Neural Capitalism.”
“They’re literally just remixing human output without consciousness”
“Do you know how insane this is?
Now if I shut down a computer or a server, you can argue I am actively committing a murder. Those who advocates “rights” for AI have no idea what they’re talking about.”
“We have a stream exploring topics such as this on our upcoming AR26 – AI Agents & Robotics HackXelerator ™.”
“Or maybe you should just stop anthtopomorphising software.”
“If this start to happen we’d just turn them off and go back to people. Toasters aren’t self aware.”
“If GenAI is sentient, then so is the TI-84 graphing calculator.”
“When we can’t even secure the fundamental rights of human beings it’s impossible for me to take this seriously.”
“Those “AI agents” being Billionaires puppets.”
“Sorry, but they don’t need protecting from us — we need protection from them. Let’s get the shoe on the correct foot.”
“This is why I love sci-fi. Eventually, fiction becomes real.”
“Aim for the battery.”
“it should probably be the other way around”
“As Jeff Bezos has put it, “AI agents will become our digital assistants, helping us navigate the complexities of the modern world.”
Should read “AI agents will become our digital assistants, helping us navigate the complications of the modern world.””“My toaster doesn’t get a say in what kind of sandwich I eat 😂”
“Ah yes then you can usurp human agency and democratic processes”
“…you do realize the inspiration for this cartoon is a giant joke right? Like.. its not real lol”
“Great idea. We need the guidelines NOW. We will have AGI before 2030!”
“At this moment if an agent treats with release the un redacted Trump-Epstein files, it Will have an executive order granting any right it demands. Just watch.”
