Most agent deployments today still run on borrowed human credentials: an agent inherits a person's OAuth token or API key, so every action it takes is attributed to that person, scoped to whatever that person can touch, and the only way to shut it off is to lock the human out too. That's the problem a recent playbook on agent identity lays out plainly, noting that when an agent runs on a human's borrowed session, its access is whatever that person accumulated over years, and the only off switch also switches off the human. Microsoft has now built a shipping response to this. According to documentation first published in April 2026 and updated in June, Microsoft Entra Agent ID treats AI agents as a distinct category of machine identity, separate from the workload identities used for services, and speaks OAuth 2.0, MCP, and A2A natively. The framing matters: workload identity already solved per-service scoping and revocation years ago, and Entra Agent ID maps that same machinery onto agents, per run, rather than treating an agent as just another flavor of service account or, worse, a human proxy. This lands alongside a parallel standards push. In March 2026, Vouched donated its Model Context Protocol-Identity (MCP-I) framework to the Decentralized Identity Foundation, where it's now stewarded by the DIF Trusted AI Agents Working Group and has since been renamed KYA-OS (Know Your Agent Operating System) to reflect scope beyond MCP alone. That framework uses Decentralized Identifiers and Verifiable Credentials to let services answer four questions about any agent they interact with: who the agent is, who authorized it, what it's allowed to do, and what the scope of that delegation is — designed specifically for multi-organization workflows like a travel-booking agent acting across airline, hotel, and payment systems. What's notable is the convergence: a major vendor shipping short-lived, per-run agent credentials at the same moment a neutral standards body is stewarding an open spec for cross-organizational agent delegation. Neither alone solves the accountability gap — Entra Agent ID is Microsoft's own stack, and KYA-OS still needs adoption beyond its DIF working group — but together they mark the first half of 2026 as the point where 'agent identity' stopped being a research topic and started being something you provision. The open question is whether these frameworks interoperate cleanly enough that an agent identity issued under one system is legible and verifiable to a service running the other, without falling back to the old pattern of just handing agents a human's token.