Singapore's IMDA released a Model AI Governance Framework for Agentic AI at Davos in January, and it's the first governance model anywhere built specifically for autonomous agents rather than AI systems generally. The core identity provision is simple to state: agents should have unique identities tied to a supervising agent or human user, so that when something goes wrong there's a traceable chain of who authorized what. That's the part worth sitting with, because the framework stops right there. IMDA organized its guidance around four dimensions — assessing and bounding risk upfront, defining human accountability and oversight, implementing technical and organizational safeguards, and enabling end-user responsibility. Agent identity falls under the accountability dimension, and the mechanism is entirely administrative: assign an identity, link it to a person, done. There's no specification for what that identity is made of — no credential format, no cryptographic binding, no revocation process. It reads more like an org chart requirement than an identity protocol. The framework is also explicitly non-binding. It doesn't create legally enforceable obligations; it's a benchmark organizations can choose to adopt or ignore. IMDA updated it again in May, which suggests the agency expects it to keep evolving rather than freezing into a standard. Compare this to what NIST is doing in parallel. A recent academic paper on agent trust protocols notes that NIST's Center for AI Standards and Innovation launched its own AI Agent Standards Initiative in February, and frames the underlying problem bluntly: agents are commonly treated as generic service accounts, without dedicated identity, authorization, or accountability infrastructure. That's the gap Singapore's framework gestures at but doesn't close. Saying an agent needs a traceable identity linked to a person is a governance requirement. Building the actual mechanism — a verifiable credential, a signed attestation of delegation, something a third party could check without trusting the deploying organization's word for it — is an engineering problem, and it's one Singapore left for someone else. This split matters because it's becoming the pattern across jurisdictions: policy bodies naming identity and accountability as requirements, technical bodies left to work out how those requirements get implemented in code. The risk is that the two tracks drift apart — regulators satisfied that a framework exists, while the actual agents running in production still look, from the outside, like unmarked service accounts with no verifiable link to anyone.