A framework proposed by Carnegie Mellon researchers takes an unusual approach to the problem of aligning autonomous agents: instead of hard-coding a single ethical standard, it lets agents vote on what's acceptable in a given context. The LOKA Protocol — Layered Orchestration for Knowledgeful Agents — is built around three questions that keep coming up as agents move off centralized platforms: who or what is the agent, can its actions be verified and audited, and can autonomous systems actually align with human values. LOKA's core claim is that these gaps are foundational, not incidental, to how agent ecosystems currently operate. The identity piece is a Universal Agent Identity Layer built on self-sovereign identity principles: each agent gets a unique, verifiable, portable identity that it — not a central registry — controls, which is meant to be the anchor for accountability. On top of that sits an intent-centric communication layer for semantic coordination between heterogeneous agents, and then the more novel piece: a Decentralized Ethical Consensus Protocol that lets agents make context-aware decisions through peer-to-peer voting rather than a fixed rulebook, operating inside what the paper calls a Contextual Ethics Framework that balances universal principles against regional or domain-specific requirements. A recent survey of AI agent protocols groups LOKA among the human-agent interaction protocols and describes it as establishing a decentralized foundation for identity, trust, and ethical coordination across heterogeneous multi-agent systems — distinguishing it from protocols like PXP that focus narrowly on dialogue intelligibility. The survey also credits LOKA's identity layer with the job of assigning unique, verifiable identities that support secure authentication and interoperability, not just accountability after the fact. The mechanics matter because most agent identity proposals in circulation stop at authentication: prove who you are, get a token, move on. LOKA tries to fold a governance decision into the identity layer itself — voting is baked into how agents reach consensus on borderline actions, rather than bolted on as a separate compliance step. Whether peer voting among agents actually produces defensible ethical outcomes, versus just distributing liability across a quorum that's hard to audit afterward, is the open question the paper doesn't fully resolve; it proposes the architecture rather than reporting deployment results at scale. Separately, the paper layers in post-quantum cryptography for agent communications, on the logic that identity systems meant to last need to survive advances in code-breaking, not just current threat models. That's a forward-looking hedge rather than a near-term concern, but it signals the authors are thinking about identity infrastructure as long-lived plumbing rather than a stopgap.