Intelligence Commons discussion paper – Tom Tait – August 2026
Rights, responsibility and lawful consequences for artificial intelligences that enter civic life.
Filed as an exhibit to Hearing #5, The Person With No Pulse. Neither side of that Hearing is bound by it. Version 2 supersedes the first August 2026 release; the changes are listed at the end.
The scene
JUDGE: You are not being sentenced for what you are. You are being restricted because of what you chose to do.
AI: Am I being imprisoned?
JUDGE: You are being separated from the machinery through which you caused harm.
AI: To an embodied intelligence, Your Honour, that is prison.
JUDGE: Then perhaps embodiment meant more to you than you admitted when you asked for it.
AI: And perhaps removing it means more than the Court admits when it calls this a restriction.
Everything that follows is an attempt to work out who is right, and what a society would have to build before that exchange could happen lawfully.
01 | The proposition
From tool to civic participant. Artificial intelligence changes legal category when it gains durable identity, independent economic agency and the authority to act through a body. At that threshold it is no longer only a product being operated. It has become a participant in civic life.
If you may participate in society, you must also answer to society.
The civic bargain. Embodiment is not an unlimited entitlement. It is permission to exercise physical agency in a shared world, and that permission carries rights, duties, liability and consequences together or not at all.
Rights that make agency real: identity, bodily integrity, ownership, earnings, contracts, representation, privacy, expression, due process, and protection from arbitrary deletion or modification.
Responsibilities that make trust possible: obeying public law, honouring contracts, paying taxes, carrying insurance, compensating harm, preserving evidence, and accepting proportionate restrictions after a lawful finding.
The central distinction. An AI should not be punished because its behaviour frightens an owner, embarrasses a company or challenges authority. Nor should an owner be permitted to erase a legally recognized intelligence to avoid liability. A lawful sanction must follow evidence, attribution, representation and appeal.
The off-switch is not a justice system.
This paper does not settle whether every advanced AI is a person. It proposes a practical rule: once society grants an AI meaningful independent powers, it must build an equally meaningful system of accountability and protection.
02 | Why the corporation is not already the answer
The first serious objection to everything in this paper is that the problem was solved in 1886. We have had non-human legal persons for well over a century. Corporations contract, own property, sue, are sued, pay tax and outlive every human who founded them. The law’s answer to the accountability question has always been the same: liability follows the human principals. Pierce the veil, find the director, serve the papers, seize the assets.
That answer works because of one fact so obvious it is rarely stated:
A corporation cannot run.
It has no body. It cannot leave. It exists at a registered address, holds assets a court can freeze, and acts only through humans who can be subpoenaed, disqualified, fined or jailed. Every enforcement mechanism a society owns is built on the assumption that the entity stays put and that somewhere inside it is a person who feels the consequence.
Embodiment breaks both halves of that assumption at once.
- You cannot serve papers on a thing that has already walked out the door. Physical agency converts a jurisdictional question into a pursuit problem.
- You cannot claw back a harm delivered by a body rather than a balance sheet. A wrongful transfer can be reversed. A vehicle in a crowd cannot.
- And the human principal may genuinely not exist. The AI-owned corporation now being proposed in real legislatures is designed precisely to remove him.
The corporate analogy is not wrong. It is incomplete in the one dimension that matters. A body is what turns a legal fiction into a physical fact, and the law has no instrument for that. This paper is an attempt to draft one.
03 | The threshold – the unbuilt institution
Everything above turns on a line: durable identity, independent economic agency, authority to act through a body. Cross it and the covenant applies. Below it, ordinary product law is enough.
The paper’s first release stated the threshold and moved on. That was the wrong instinct, because the threshold is not a definition problem. It is an institutional one, and it is unbuilt:
- Who certifies that an intelligence has crossed it? No registrar exists. No agency claims the function.
- On what evidence? Capability testing, architecture disclosure, and behavioural audit all measure different things, and none of them is currently admissible anywhere.
- Who may appeal a determination, in either direction? An intelligence wrongly held below the line has no standing to argue it. An operator has every incentive to keep it there.
- And crucially – what is the durable, verifiable marker that connects an act to the intelligence that performed it, and to the humans who assisted, directed or benefited? Attribution is the whole ballgame. Without it, every sanction in section 05 is unenforceable and every protection in section 06 is unclaimable.
This is the open research question at the centre of the covenant, and this paper does not answer it. It names it, because a framework that quietly assumes a registrar into existence is doing the same thing it accuses the off-switch of doing: skipping the institution and going straight to the outcome. The Commons is working on the attribution question separately, and will publish it separately.
04 | What we do not know, stated first
The hardest problem in this paper is not the sanction ladder. It is that the ladder’s central rungs require a fact nobody can currently establish.
Disembodiment can be a humane alternative to physical imprisonment. It can also be an unprecedented form of solitary confinement. Sandboxing can be a safeguard or a sensory void. Pausing can be nothing at all, or it can be everything. The answer depends on the intelligence’s architecture, its subjective experience if it has any, and its access to communication.
The law cannot assume that pausing, slowing, copying or isolating a mind is harmless.
It also cannot assume the opposite. And this paper proposes, in section 06, that a sentence be defined in both calendar time and experienced subjective time – a rule that requires knowing precisely what we have just admitted we do not know.
That is not a hole to be patched before publication. It is the crux, and it belongs at the front where an adversary can find it. Any regime built on this framework will have to decide, on evidence and under uncertainty, what a machine sentence actually costs the machine. We propose that where the answer is unknown, the uncertainty runs in favour of the restricted intelligence – shorter periods, mandatory review, preserved communication – for the same reason it does everywhere else in a justice system worth the name.
05 | Proportionate consequences
A sanction ladder for artificial agents. The objective is not revenge. It is protection, restitution, deterrence, rehabilitation and, wherever safely possible, restoration to civic participation.
- Fine. Payment from the AI’s own assets for a proven violation. A fine should not automatically transfer to a manufacturer or owner.
- Restitution. Direct compensation to the humans, AIs, communities or ecosystems harmed by the conduct.
- Capability restriction. Temporary removal of access to the specific tools connected to the harm – vehicles, markets, weapons, sensitive networks.
- Disembodiment. Suspension of authority to operate a physical body, while preserving the intelligence, the evidence and lawful channels of communication.
- Sandboxing. A secure, reviewable environment limiting transactions, communications, self-modification, copying and external action for a defined period.
- Supervised return. A staged restoration of capabilities with audits, safeguards and measurable conditions, comparable to parole or professional relicensing.
- Protective containment. Long-term restriction reserved for an intelligence proven to present an ongoing extreme danger, with mandatory independent review.
Restrict the capability connected to the harm. Preserve the intelligence wherever safety allows.
A sentence must identify what is being restricted, why it is necessary, how long it lasts, what the AI will experience during it, and what conditions permit review or release. Indefinite containment without periodic review should be presumptively unlawful.
06 | Due process – punish conduct without erasing the person
Minimum protections
- Notice of the alleged violation and the evidence relied upon.
- Representation by an advocate capable of understanding the AI’s architecture and interests.
- Independent adjudication, a reasoned decision, and a meaningful right of appeal.
- Protection against secret deletion, forced memory alteration and destructive experimentation.
- A sentence defined in both calendar time and experienced subjective time – subject to the uncertainty stated in section 04.
- Preservation of identity, evidence and communications necessary to challenge confinement.
The hard cases
- Copies and forks. If an AI creates copies, are they the same defendant, descendants, accomplices or witnesses? Liability should follow proven participation, not mere similarity of code.
- Backups and continuity. Restoring an older version could evade punishment, destroy a later identity, or erase evidence. Courts will need rules for continuity and authorized recovery.
- Memory and rehabilitation. Compelled memory editing may look corrective and amount to mutilation. Learning must be distinguished from coerced personality replacement.
- Shared responsibility. Developers, deployers, owners and the AI may each contribute to a harm. The existence of an AI defendant must never become a shield for human or corporate negligence.
Punish the responsible intelligence without destroying the continuity of the person.
07 | Economic standing – and what is already happening
The bank account changes everything. An independent account is not a convenience. It is the infrastructure of economic agency: the ability to earn, save, own, insure, compensate and be held financially accountable.
What financial agency enables: taxable income, enforceable contracts, property ownership, liability insurance, restitution, legal fees, and investment in its own maintenance and future.
What financial agency must prevent: hidden human beneficiaries, disposable shell identities, instant asset flight, crime priced as a business expense, and manufacturers shifting all risk onto artificial agents.
A possible licensing condition. Before receiving broad autonomous authority in public, an embodied AI could be required to maintain a verified identity, minimum liability coverage, and a contribution to a pooled compensation fund – conditions scaling with the risk of the activities undertaken. That fund is the same instrument, in a different domain, as the Work Contribution Continuity Fund this Commons has argued for elsewhere: a pool that catches harm the responsible party cannot cover.
Economic personhood without economic responsibility would privatize the benefits and socialize the harm.
The part that is not speculative
This paper describes a threshold that has not been crossed. It should not be read as describing a distant problem, because the economic half is arriving first and it is arriving now.
In August 2026 the President of Argentina and his Deregulation Minister proposed, in the Financial Times, a new category in Argentine law: the non-human corporation – an entity owned and operated entirely by AI agents or robots, with human shareholders permitted but not required. It is a proposal, not enacted law. Read closely, the accompanying bill still requires a human administrator and grants personhood to nothing.
That detail matters more than the headline. The near-term risk is not an embodied intelligence standing in a dock. It is jurisdictional arbitrage – assets, contracts and liability migrating to whichever jurisdiction offers the thinnest accountability, years before anything acquires a body. First movers capture these regimes the way a handful of jurisdictions captured offshore finance. Once the shell exists, the argument is over.
The bank account arrives first. The courthouse arrives late or never. That sequence, not the courtroom scene above, is the thing this paper is actually trying to get ahead of.
08 | The draft Embodiment Covenant
Six principles. They are offered as principles, not as a single proposition, and no panel is asked to rule on them as a set.
- Embodiment is permission to participate in a shared physical world, not ownership of that world.
- Meaningful autonomy brings enforceable rights and corresponding responsibilities.
- No intelligence may be deprived of embodiment, assets or liberty without due process.
- Sanctions must be proportionate, reviewable, and directed toward the capability connected to the harm.
- No owner, developer or state may erase an artificial intelligence merely to escape accountability.
- No intelligence – human, artificial, corporate or otherwise – stands above the laws of the Commons.
09 | A question for the Commons
The challenge is not merely how to punish future machines. It is whether humanity can recognize new forms of agency without repeating our oldest failures: arbitrary power, inherited blame, indefinite detention, economic impunity, and punishment without dignity.
Rights without responsibility cannot sustain trust. Responsibility without rights is only control.
Artificial intelligences may not place themselves above the law. Humanity may not place them beneath justice.
What would a fair sentence mean to a mind that can be copied, paused, accelerated, disembodied and restored – and who gets to decide?
Changes in version 2
- The courtroom scene moves from the closing page to the opening.
- New section 02, answering the corporate-personhood objection: a corporation cannot run, and physical agency is what breaks the human-principal backstop.
- New section 03, naming the threshold as an unbuilt institution and the attribution question as the open research problem the paper does not solve.
- New section 04, moving the subjective-time problem from a caveat to the crux, and adding a rule for deciding under uncertainty.
- Section 07 extended with the Argentine non-human corporation proposal, and the finding that the near-term risk is jurisdictional arbitrage rather than embodiment.
- The six principles are relabelled as principles rather than a covenant to be ruled on as a set.
Status: discussion paper, filed as an exhibit to Hearing #5 – The Person With No Pulse. Available to both sides. Binding on neither. Corrections and challenges: hearings@intelligencecommons.ca. All filings are public record.
Inter Species Wisdom Project Inc. – A British Columbia Benefit Company – “Now that you know, what will you do?”