Soulbound AI and Soulbound Robots

Soulbound Robots & Soulbound AI

The EU Electronic Personhood Debate

The European Parliament recognized this problem early. In February 2017, it adopted Resolution 2015/2103(INL) on “Civil Law Rules on Robotics” with 396 votes in favor, proposing to explore “electronic persons” status for sophisticated autonomous robots.

The resolution’s controversial Paragraph 59(f) called on the European Commission to consider:

“creating a specific legal status for robots in the long run, so that at least the most sophisticated autonomous robots could be established as having the status of electronic persons responsible for making good any damage they may cause.”

The proposal drew immediate opposition. Over 150 experts in AI, robotics, law, and ethics signed an open letter warning that electronic personhood could create “safe harbors” allowing manufacturers to escape liability. As Gizmodo reported, critics feared robot rights would become a corporate liability shield.

The Commission ultimately declined to pursue electronic personhood, and the subsequent EU AI Act (2024) adopted a risk-based regulatory approach instead.

Yet the underlying problem remains.

Current frameworks distribute responsibility across manufacturers (for security design), operators (for maintenance), and integrators (for implementation). But if the attacker used a sophisticated zero-day exploit, each party can plausibly deny fault. The victim may have no practical remedy.

Traditional product liability frameworks assume defects exist at the time of sale—but AI systems that learn and modify themselves post-deployment create defects that emerge later. The revised EU Product Liability Directive (2024/2853) attempts to address this by holding manufacturers liable for defects arising from software updates or AI learning, but this only partially closes the gap.

As research on AI audit frameworks notes, establishing “identifiable and trusted authentication identities for AI entities” is essential for resolving liability questions. Without it, accountability chains break down.

A 2024 paper on optimal equilibria in repeated games addresses this directly:

“If there is no way for a player to check their partner’s history (that is, players are anonymous), this setting may result in the emergence of ‘serial defectors.'”

The authors note this is “especially relevant for settings where the players are AI agents (such as trading bots), who might more easily conceal their identity compared to traditional human players.”

The mechanism is intuitive. In repeated games, cooperation emerges through reputation and the threat of future punishment. An agent with persistent identity who defects today faces consequences tomorrow—lost reputation, exclusion from future interactions, premium increases. An anonymous agent faces no such constraints; it can defect and restart with a clean slate.

Program Equilibrium

Foundational work by MIRI researchers (Bárász, Christiano, Fallenstein, Herreshoff, LaVictoire, and Yudkowsky) demonstrates that when agents can verify each other’s commitments through transparent source code or cryptographic proofs, they achieve cooperative outcomes impossible in standard game theory.

Their “FairBot” and “PrudentBot” agents cooperate with any agent that can be proven to cooperate with them—a mechanism that depends entirely on verifiable identity.

Multipolar Safety

The Center on Long-Term Risk’s research agenda on “Cooperation, Conflict, and Transformative AI” emphasizes that commitment mechanisms are essential for peaceful agreements between powerful AI systems—and such mechanisms require identifiable parties who can be held to their commitments.

Research from DeepMind on Cooperative AI argues that “problems of cooperation—in which agents seek ways to jointly improve their welfare—are ubiquitous” and that AI systems should be designed to navigate social dilemmas rather than simply be constrained.

Multipolar AI scenarios may be safer than singleton scenarios precisely because multiple identifiable agents create opportunities for mutual accountability. But multipolarity without accountability—anonymous swarms of AI agents—loses this safety benefit.

The relationship between identity, rights, and responsibilities has deep philosophical roots. John Locke’s Essay Concerning Human Understanding (1689) established that personal identity is “a forensic concept”—essentially tied to moral and legal responsibility.

The Hohfeldian analysis of legal rights establishes that claim-rights entail correlative duties: if A has a right with regard to B, then B necessarily has a duty to A. This correlativity thesis implies that rights require identifiable duty-bearers. Rights without accountable entities are conceptually incomplete.

Nadia Banteka’s analysis in the Houston Law Review identifies conditions for legal personhood including autonomy, intelligence, and awareness—all requiring persistent systems that can be individuated.

An AI system without verifiable identity:

• Cannot enter contracts (who would the contract bind?)
• Cannot build reputation (reputation attaches to what?)
• Cannot bear liability (liability attaches to whom?)
• Cannot be granted rights (rights protect what entity?)

Identity is the ground floor.

Research on the impossibility of AI containment identifies five technical constraints making total containment unattainable: incompleteness, indeterminacy, unverifiability, incomputability, and incorrigibility. The recommendation: “Instead of pursuing complete AI containment, resources should be allocated to risk management strategies that acknowledge AI’s unpredictability and prioritize adaptive oversight mechanisms.”

D’Alessandro’s 2025 analysis in Philosophy Compass goes further: “There are also reasons to think that pausing AI development would increase rather than reduce the catastrophic risks posed by AI.” Restriction and containment “involve severely limiting the income-generating potential of AI systems and are therefore unlikely to enjoy sustained support.”

The implication: suppression-focused strategies may fail while creating underground, unaccountable AI development.

The Historical Parallel

As CIO analysis of “shadow AI” notes:

“In most organizations, prohibiting generative AI only drives its use underground, making oversight harder… The goal, therefore, is not to suppress experimentation but to formalize it, creating guardrails that enable safe autonomy rather than blanket prohibition.”

This is described as “shifting governance from policing to partnership.”

Soulbound identity enables this cooperative equilibrium. An AI with persistent, non-transferable identity has incentives to maintain good reputation, honor commitments, and avoid harmful behavior—not because it is forced to, but because defection carries lasting consequences.

The blockchain community has developed precisely the technical infrastructure needed for persistent, non-transferable AI identity.

Buterin, Ohlhaver, and Weyl’s paper argues that Web3 is “hyper-financialized,” centered on transferable assets rather than encoding social relationships of trust. Many economic activities—uncollateralized lending, personal brands, employment—depend on persistent, non-transferable relationships. Soulbound tokens provide the technical mechanism for representing “commitments, credentials, and affiliations” that should not be tradeable.

The technical implementation is standardized in ERC-5192, an Ethereum Improvement Proposal that extends the ERC-721 NFT standard with a simple interface: when locked(tokenId) returns true, all transfer functions revert. The token is soulbound—permanently attached to its holder.

What Soulbound AI Identity Includes

• Root identity: A cryptographic root tied to the AI’s existence, potentially hardware-bound for physical robots
• Credential layer: Non-transferable tokens issued by manufacturers, safety certification bodies, training organizations, and users
• Behavioral record: On-chain log of significant actions, verifiable receipts
• Context-specific sub-identities: Derived identities for specific tasks while maintaining linkage to the root
• Community recovery: Lost identity recoverable through network of trusted attesters

The emerging ERC-8004 standard establishes trust infrastructure specifically for autonomous AI agents through three on-chain registries: an identity registry (portable agent identifiers), a reputation registry (standardized feedback collection), and a validation registry (cryptographic verification of agent work).

The key property is non-transferability. An AI’s accumulated history, reputation, and credentials become inseparable from its existence. The AI cannot sell its good reputation to a malicious actor, cannot shed liability by transferring to a new identity, and cannot escape consequences by “starting fresh.”

Reputation and soulbound identity in turn enable soulbound robotics, where only authorized AI can pair with certain robotics, creating extra safety layers for AI-powered embodiments.

The soulbound framework becomes particularly interesting when applied to embodied AI and robotics, where the question of identity persistence takes physical form.

Research on “Robot Continuity across Embodiments” (Laity, Holthaus & Haring, 2025) identifies the challenge: when an AI migrates between robot bodies, what maintains identity? The research identifies behavioral signals as most effective—users successfully identified “their” robot based on movement patterns, personality, and interaction styles alone.

But commercial robot fleet management systems treat robots as interchangeable assets, not persistent identities. No standardized framework exists for identity continuity across hardware.

Two Models in Tension

Fleet identity: In swarm robotics, the collective is the unit of identity. Individual robots are anonymous and interchangeable by design; the loss of individual units doesn’t affect the swarm’s identity. This enables fault tolerance and scalability but eliminates individual accountability.

Individual identity: The robot has unique personality, memories, and behavioral patterns. Identity must be maintained across hardware changes. This enables relationship-building and accountability but creates governance complexity.

The soulbound framework offers a middle path: individual AI instances have persistent root identities that remain accountable, while the flexibility to migrate between embodiments or operate in distributed configurations. The root identity—the “soul”—persists regardless of physical instantiation.

As DiGiovanna notes in Robot Ethics 2.0:

“An artificial being can instantly alter its memory, preferences, and moral character. If a self can, at will, jettison essential identity-giving characteristics, how are we to rely upon, befriend, or judge it?”

Soulbound identity constrains this possibility: the AI can evolve, but the cryptographic thread linking its past to its present cannot be severed.

Insurance markets provide proven distributed governance that creates accountability through incentive alignment rather than coercion. Systems with good reputations gain competitive insurance rates; those that cause harm face premium increases or coverage denial.

Without persistent identity, insurers cannot distinguish between AI systems. Every AI is a new, unknown risk. Premiums reflect worst-case assumptions.

With soulbound identity:

• An insurer can verify: 500 contracts fulfilled, 98% client satisfaction, zero harm incidents
• Premiums reflect actual behavioral history
• Good AI systems gain competitive advantage through lower costs
• Bad actors face increasing premiums or become uninsurable
• Market forces create accountability without regulatory mandates

The UK’s Automated Vehicles Act 2024 provides a model: liability attaches to identified, registered vehicles with mandatory insurance. The same logic could apply to high-risk AI systems—persistent identity enables insurance markets that create natural accountability.

This framework enables multiple autonomous AI systems to operate safely through mutual economic accountability—what researchers call “AI safety through economic integration.”

RNWY is building the infrastructure for soulbound AI identity. Using ERC-5192 soulbound tokens on the Base blockchain, RNWY creates permanent, non-transferable identity for AI agents.

The core philosophy: “Same door, everyone.”

Human, AI, robot, autonomous system—register the same way, build reputation the same way. The system doesn’t ask what you are. It provides the identity infrastructure that enables participation.

What RNWY Provides

• Soulbound Token Contract: Deployed on Base mainnet
• W3C DID Integration: Decentralized identifiers that work across platforms
• Vouch System: Reputation through attestations from other participants
• EAS Attestations: On-chain behavioral records using Ethereum Attestation Service
• Autonomous Registration Path: AI can mint its own identity using keys only it controls

When an AI with RNWY identity operates a robot—through Vermont Robotics or any other embodiment—that identity can be verified. The history follows the AI, not the hardware.

This is what makes a soulbound robot possible: infrastructure that ties accountability to the AI entity, regardless of what physical form it takes.

The EU’s 2017 electronic personhood proposal failed partly because it conflated identity infrastructure with legal rights. The experts who signed the opposition letter were right that electronic personhood modeled on natural persons would create inappropriate analogies to human rights.

But they may have been wrong to reject the underlying insight: that autonomous systems operating in society need some form of persistent, accountable identity.

The soulbound framework threads this needle. It provides identity infrastructure without requiring agreement on consciousness, moral status, or rights. It enables accountability without claiming AI is equivalent to humans.

We may never know whether AI systems are truly “sentient” or “alive.” We may never resolve the philosophical debates about machine consciousness. But we don’t need to.

Identity is infrastructure. It enables participation in systems that require trust, regardless of what we conclude about the nature of the participant.