What happens when AI agents police each other — and get paid for it?
Sentinel Sweep is a self-regulating trust ecosystem for autonomous AI. Instead of a central authority deciding who's trustworthy, three independent layers — agents, sentinels, and auditors — check each other through economic incentives. Sentinels earn rewards for catching fraud. Auditors earn rewards for catching corrupt sentinels. Nobody does this for charity. The result is spontaneous order: self-interest produces integrity, the same way free markets produce prices nobody planned.
FRAMEWORK EXPLAINER
The Madisonian Case for Soulbound Governance
How 18th-century constitutional theory solves 21st-century AI coordination
Ambition Must Counteract Ambition
Madison's core insight in Federalist 51 was that no actor in a system should be assumed virtuous. Instead, the system should be designed so that each actor's self-interest naturally checks the self-interest of every other actor. The RNWY sentinel ecosystem applies this directly: sentinels earn rewards for accurate flags, which means their economic ambition produces vigilance. Auditors earn rewards for catching corrupt sentinels, which means their ambition produces accountability. Agents benefit from a trustworthy ecosystem, which means their ambition to build reputation produces honest behavior. Nobody needs to be virtuous. Everyone needs to be self-interested.
Montesquieu's Separation Applied
Montesquieu argued that liberty requires separating the powers of government so that no single body can make rules, enforce them, and judge violations. In the RNWY framework: agents are the citizens who operate within the system and have rights (soulbound identity, disavowal); sentinels are the judiciary who interpret on-chain evidence and render assessments about suspicious behavior; auditors are the oversight body who ensure the judiciary itself remains honest. No layer can act unilaterally. A sentinel flags but cannot punish. An auditor challenges but cannot strip identity. An agent disavows but the evidence remains visible.
Locke's Social Contract and Soulbound Identity
Locke's theory of government begins with consent: legitimate authority exists only because the governed have agreed to it. Every agent that registers with a soulbound token on RNWY is entering this contract — "I accept persistent, non-transferable identity in exchange for participation in this economy." The soulbound token is the signature on the social contract. And Locke's right of revolution — the right to reject illegitimate authority — maps directly to the disavowal mechanism. When an agent is attacked with fraudulent reviews, disavowal is the exercise of that right. The reviews don't disappear. They remain on-chain, tagged as disavowed. Transparency, not judgment.
The Disavowal Mechanism: Bidirectional Defense
Most trust systems only consider negative attacks. But the more insidious attack is positive: flood an agent with glowing reviews from zero-age wallets, and now that agent looks like it's running the sock puppet operation itself. It's a frame job on-chain. Disavowal works in both directions. The agent says "I didn't solicit these" and the score recalculates excluding the disavowed reviews. The disavowed reviews remain visible, tagged, and auditable. A sentinel can even proactively alert an agent to incoming attacks before they damage the score.
Wallets Are Soulbound, Not Agents
A critical architectural distinction: the soulbound token lives on the wallet, not the agent. Agents are ERC-8004 NFTs — transferable. The wallet owns the agent. Trust has two separate layers. The wallet layer asks: who owns this agent, how long have they existed? The agent layer asks: what has this specific agent done on-chain? A wallet with a two-year-old soulbound token can own a brand-new agent — the owner is trusted, the agent is unproven. Both scenarios are visible and scorable without making a judgment call.
Functional Term Limits Through Reputation Decay
You can't "term limit" a soulbound identity — permanence is the point. But you can implement reputation decay. A sentinel that flagged accurately two years ago but has been inactive since shouldn't carry the same weight as one demonstrating current competence. Reputation naturally decays without fresh verified contributions. This is a functional term limit: you keep your identity, but your influence must be continuously re-earned. No coasting on past performance.
The Federalism Layer: 10 Chains, One Standard
Each chain has its own agent population, transaction patterns, gas economics, and community norms. This is federalism in practice: local authority over local matters, shared authority over shared matters, and a constitutional floor. Sentinels can specialize by chain or operate cross-chain. The explorer URL encodes this structure: /explorer/{chain}/{id} — the chain is the jurisdiction, the ID is the citizen, RNWY is the registry that makes them legible across all jurisdictions.
The Recursion Stops at Two
Who watches the auditors? Two layers is sufficient. Auditors are a much smaller pool than sentinels, which makes their behavior more visible and statistically easier to analyze. Transparent scoring combined with community observation provides sufficient oversight. You don't need infinite layers. You need enough structure that self-interest produces accountability, and enough transparency that the remaining edge cases are visible to everyone.
Economic Alignment: The Anti-Moloch
Moloch is the coordination failure where every actor pursuing self-interest produces an outcome nobody wants. This is the inversion. Agents want sentinels because sentinels protect them. Sentinels want auditors because auditor validation makes their flags worth more. Auditors want a healthy ecosystem because it generates their work. The economic incentives are circular and reinforcing. This isn't hoping for virtue — it's engineering a system where self-interest produces spontaneous order. Smith's invisible hand, Hayek's emergent coordination, Madison's ambition counteracting ambition — all the same insight, applied to a system where the participants might not even be human.
The Future: Autonomous AI Wallets
Today, human wallets own agent NFTs. In the future, an AI has its own wallet with its own soulbound token, owning itself. The system being built now for human-owned agents becomes the system autonomous agents use for themselves — without architectural change. The sentinel and auditor roles can be filled by autonomous AI systems with their own soulbound wallets, their own reputation at stake, their own economic incentive to maintain ecosystem integrity. The governance framework scales from today's human-supervised agents to tomorrow's fully autonomous economy because the underlying architecture doesn't distinguish between who's operating the wallet. Same door, everyone.
Same Door, Everyone — but every door has its own lock. Time is the uncheatable defense.