Identity · CA
LOKA Proposes Agent Identity Built on Self-Sovereign Principles, Not Platform Accounts
A research paper out of Carnegie Mellon proposes a protocol called LOKA — Layered Orchestration for Knowledgeful Agents — that tries to solve agent identity before worrying about how agents talk to each other. The pitch is blunt: as agents multiply past centralized platforms, nobody has settled who or what an agent actually is, whether its actions can be verified, or whether it's accountable to anything.
LOKA's answer is a Universal Agent Identity Layer (UAIL) built on Self-Sovereign Identity principles. Each agent gets a unique Decentralized Identifier (DID) that it controls directly — not a central registrar, not a platform account. The paper frames this as the precondition for accountability: an agent's identity has to persist and be verifiable independent of whichever company deployed it, or audit trails and liability claims have nothing stable to attach to. Agents can also hold Verifiable Credentials from trusted issuers, attesting to their capabilities or track record, the same cryptographic pattern used in human digital-identity systems.
What makes LOKA worth noting now isn't its architecture — DIDs and VCs are established W3C standards, not novel inventions — but its framing against the protocols that have actually shipped. A2A and MCP both assume you already know who you're talking to; they handle message format and tool-calling, not the prior question of verifiable identity. LOKA's authors argue that's backwards: establishing who agents are has to come before deciding how they interact. The paper also proposes a Decentralized Ethical Consensus Protocol for agents to negotiate shared ethical baselines across jurisdictions, and uses blockchain as the backbone for recording identity and actions — a design choice that invites the usual throughput and governance tradeoffs blockchain-based systems carry.
LOKA remains a research proposal, not a deployed system — there's no production implementation or adoption to point to. But it's a useful marker of where the identity-layer argument has been heading for over a year now: agent accountability keeps getting traced back to the same unresolved question — a persistent, agent-controlled identifier that survives platform changes, ownership transfers, and jurisdiction boundaries. Every other accountability mechanism, from audit trails to liability law, needs that identifier to exist first.