A protocol proposal out of Carnegie Mellon is trying to solve agent identity and agent ethics with the same architecture, rather than treating them as separate problems. The LOKA Protocol — Layered Orchestration for Knowledgeful Agents — frames the challenge around three questions that most current agent stacks answer only partially or not at all. The paper's framing is blunt: as autonomous agents proliferate beyond centralized infrastructure, three critical questions emerge — identity, accountability, and whether these systems can reliably align with human values and avoid harmful emergent behavior. The identity piece looks familiar to anyone tracking this space: a Universal Agent Identity Layer built on self-sovereign identity principles, assigning each agent a decentralized identifier plus verifiable credentials attesting to its capabilities and ethical standing. Verifiable credentials here work on an issuer-subject-verifier model, where a trusted organization signs a credential for an agent, and any other agent or human can check it independently — the same pattern showing up across DID-based agent identity proposals generally. What's less common is what LOKA bolts onto that identity layer: a Decentralized Ethical Consensus Protocol that lets agents make collaborative ethical decisions through peer-to-peer voting mechanisms, operating inside a Contextual Ethics Framework meant to balance universal principles against regional or domain-specific rules. Rather than an external auditor checking agent behavior after the fact, the idea is that a network of agents with verified identities can vote on borderline decisions in real time, with delegated authority going to agents whose reputation and track record earn it. It's worth being skeptical of how this would actually work at scale — peer voting among software agents raises its own questions about collusion, gaming, and who counts as a legitimate voter in the first place. The paper itself is a proposal, not a deployed system, and comes from an academic group rather than an implementation with production traffic behind it. Coverage of the paper has noted it positions itself alongside existing coordination standards like Google's Agent2Agent and Anthropic's Model Context Protocol rather than replacing either. The interesting part isn't the identity layer — DIDs and verifiable credentials for agents are becoming standard vocabulary. It's the attempt to fuse that identity layer directly with a governance mechanism, betting that accountability and ethical alignment can't be bolted on separately after agents already have verified identities.