NIST's Center for AI Standards and Innovation launched its AI Agent Standards Initiative in February, organized around three pillars: industry-led standards, open-source protocol development, and security-and-identity research. The most concrete artifact so far is a companion NCCoE concept paper, "Accelerating the Adoption of Software and AI Agent Identity and Authorization," which proposes adapting existing identity protocols — OAuth 2.0/2.1, OIDC, SPIFFE/SPIRE — for non-human agent workloads. What's notable isn't the protocol list. It's the specific question NIST is putting to industry: how can agents log actions in tamper-proof ways, and how do you establish non-repudiation for a decision an agent made autonomously? That's a narrower and harder problem than "give the agent an identity." An OAuth token tells you an agent was authorized to act. It doesn't tell you, after the fact and in a way a third party can verify, that the agent actually did what its logs claim, using the reasoning it claims, without the log itself having been edited or the agent having been swapped mid-session. The paper also asks how to apply least-privilege and zero-trust principles when an agent's actions aren't fully predictable in advance — a genuine departure from how these principles were designed for human or static-service principals, where the action space is bounded ahead of time. NIST's framing treats this as an open research question rather than a solved one, which is honest: no existing standard currently answers it. The process matters as much as the content. NIST ran a Request for Information on AI agent security through March 9, with comments on the identity and authorization concept paper open into April, and sector-specific listening sessions following after. Respondents to the RFI, including the Foundation for Defense of Democracies, pushed NIST to update existing federal security guidance (SP 800-160, SP 800-218) specifically for agentic systems and to extend MITRE's ATLAS framework to cover multi-agent lateral movement. The gap this exposes is structural, not just technical. Enterprises today are largely running agents on shared service accounts and API keys — credentials built for static services, not autonomous actors with unpredictable action spaces. NIST's own framing is that this arrangement doesn't scale: fragmented identity systems now carry the same credential-sprawl problems that plagued traditional IT, but at machine speed and without a settled answer for what a tamper-evident, third-party-verifiable audit trail for an autonomous decision actually looks like. That answer is what the Initiative is now trying to source from industry rather than dictate from Gaithersburg.