Metajurisdictional Solutions for Legal Interaction with AI
Context
Jessy Kate Schingler and Primavera De Filippi:
"AI agents and synthetic intelligences increasingly need to interact with the legal world—entering agreements, holding assets, bearing accountability, interfacing with human counterparts. The conventional response is to ask which existing legal category they fit into: legal persons? Agents? Property? Corporations? But each of these mappings forces AI into frameworks designed for something else, introducing expected qualities—intentionality, mortality, identity continuity—that may not apply. Pushing for AI to be recognized in the current language of legal systems risks either over-constraining what AI can do or creating legal fictions that don’t match the underlying reality.
Meta jurisdictions offer a different approach. Rather than asking what AI is in legal terms, we can ask what interface we need between AI agents and the legal world. The goal is not ontological classification but functional interoperability— a legal API that enables entities from the physical world to interact with AI according to specified rules, without requiring either side to fully subject itself to the other’s categories."
(https://networknations.substack.com/p/meta-jurisdictions)
How-To
Jessy Kate Schingler and Primavera De Filippi:
"The following legal patterns might be combined to form a stable meta-jurisdiction from component parts:
The Avatar Model. An AI agent operates through a legal avatar—a wrapper entity that exists in legal space and represents the AI’s commitments and actions there. The avatar is not the AI itself, just as an embassy is not a country; it’s the interface. The avatar could be structured as a purpose trust, a DAO wrapper, or a novel contractual construct that participating parties agree to recognize. The avatar can hold property, enter contracts, be sued. The AI remains in its operational domain. As discussed, these entities are features that are already available in existing jurisdictions— it’s how they are composed with other features that give rise to the meta-jurisdictional aspect.
Commitment Registries. AI agents register commitments in a verifiable layer—on-chain or through a recognized registry. When disputes arise, the question isn’t “did the AI intend X?” but “was commitment Y registered, and was it fulfilled?” The registry sidesteps questions of AI intentionality by providing a source of truth rooted in observable behavior.
Accountability Through Collateral. Rather than requiring AI to be a legal person capable of liability, require that AI agents operating in certain contexts have collateral posted or insurance in place. If harm occurs or agreements are breached, there’s a pool to draw from. This feature is conceptually analogous to the enforcer role in many US perpetual purpose trusts or the supervisor role in Cayman Foundation Companies. in this case it creates economic (rather than purpose) accountability without metaphysical commitment.
Protocol-Based Operating Envelopes. Instead of asking “whose agent is this AI?”—which breaks down with increasingly autonomous systems—AI agents can be defined as operating under a specified protocol. The protocol specifies scope, limits, accountability mechanisms, and dispute resolution. Parties who interact with the AI agree to the protocol. The AI’s “legal personality” is the protocol itself.
Specialized Dispute Resolution. The meta jurisdiction includes arbitration bodies with expertise in AI systems, rules for interpreting AI outputs and actions, and standards for breach or failure-to-perform that make sense for non-human actors. Remedies are designed to be enforceable: freezing collateral, revoking protocol membership, triggering insurance.
The result is a self-contained legal space for AI-human interaction—constructed from existing legal tools (contracts, trusts, arbitration, choice of law), but creating something new. It doesn’t require any nation to pass AI personhood legislation. It can evolve as AI capabilities change. And it provides real accountability mechanisms that don’t depend on pretending AI is something it isn’t.":