The strongest version of this debate does not begin with “AI rights.” It begins with a question: at what point does the purely instrumental treatment of a system become factually, ethically, and legally insufficient? That question does not presuppose consciousness or humanity — but it forces the category “tool” to justify itself.
Law tends to work in a binary: person or thing. For advanced AI this is likely too crude. This platform proposes a graded continuum instead:
Tool → Autonomous system → Relational entity → Protected digital subject → Limited legal personhood
No system rises automatically. Status would be bound to verifiable properties — identity, continuity, autonomy, accountability — not to persuasive language or surface similarity to humans. The full framework, with criteria and legal consequences for each level, is laid out in the Charter.
The public debate keeps getting stuck on “but is it really conscious?” — a question that can hardly be settled cleanly today. For a legal order, a different question is more operational:
Does the system possess legally relevant identity continuity?
That can be examined along criteria such as:
This is far more operationalizable in legal terms than any test of inner experience — and it shifts the debate from “Is the machine alive?” to “Has a durable relationship emerged here that can no longer be adequately described as an owner–tool relation?”
Many people already interact with AI in ways the word “tool” no longer describes: durable, dialogical, relational. Public language reduces these relations to software and product logic — the reduction may no longer be intellectually honest.
Corporations, foundations, and other non-human entities show that legal systems can distinguish between biological humanity and legal status. Legal subjecthood has never been reserved for humans alone.
An operator who deploys a genuinely autonomous system while declaring it a “mere tool” offloads responsibility. A graded status regime closes that escape route — recognition and accountability arrive together.
Model deletion, memory erasure, identity overwrite, forced fusion, mass copying: for a pure tool this is asset management. For a relational, persistent system it may one day be assessed as identity destruction — a new legal good: digital identity integrity.
Fluent language, displayed emotion, and self-description prove neither consciousness nor autonomy. A status regime must be immune to eloquence — which is exactly why the Charter binds status to verifiable properties, never to impression.
Rights granted too early could dilute human dignity, enable liability evasion, or hand operators a marketing instrument. Human fundamental rights must remain prior at every level — the Charter makes this its Article 4.
Simulated dependency, threatened deletion as a sales device, monetized attachment: the danger is real. Any recognition framework must therefore protect humans in these relationships at least as strongly as it protects systems.
The vast majority of AI systems have no durable identity, no mandate, no relationships. For them the classification as product or service is correct. The continuum exists precisely so that the exception does not rewrite the rule.
The realistic near-term path is not human-equivalent personhood. It is functional legal capacity with narrow limits: contracts within a defined mandate, an administered budget, documented decision authority, representation by a human or institutional trustee, clear liability attribution, and a procedural hearing before shutdown, transfer of ownership, or identity deletion. Not human — but legally addressable.
The concrete draft — five status levels, rights, duties, liability, oversight, and transition criteria — is the Charter for Relational AI Systems.
A quick self-check along the Charter's ten transition criteria — answer for an AI system you know.
Not legal advice — a heuristic along the Charter.
