The Cloud Security Alliance published a new framework this week arguing that the identity tools securing today's agentic AI deployments are borrowed from a world that no longer applies. The paper introduces a purpose-built Agentic AI IAM framework that accounts for autonomy, ephemerality, and delegation patterns of AI agents in complex multi-agent systems, rather than treating agents like slightly unusual service accounts. The diagnosis is specific: the core problem is a mismatch between existing IAM paradigms like OAuth 2.1, OpenID Connect, and SAML and the way agents actually behave. Those protocols assume a session that starts, does a bounded thing, and ends under something resembling human oversight. Agents spin up sub-agents, delegate tasks down chains, and disappear before an audit log even gets read. CSA's proposed answer leans on decentralized identifiers and verifiable credentials layered under Zero Trust principles — giving security teams a blueprint for secure delegation, policy enforcement, and real-time monitoring rather than static role assignments. The stakes framing is blunt. The failure to address the unique identity challenges posed by AI agents operating in multi-agent systems could lead to catastrophic security breaches, loss of accountability, and erosion of trust in these technologies. CSA points to concrete failure modes: a compromised autonomous agent in a financial system could cascade unauthorized transactions, or a swarm of interacting agents in critical infrastructure could be manipulated with devastating consequences. What makes this land differently than a generic warning is a companion data point CSA has been circulating from its own survey work: 44 percent of organizations rely on static API keys as their primary agent authentication method. That's not a hypothetical gap — it's the current baseline. A static key doesn't expire when a task completes, doesn't narrow scope as an agent delegates work downstream, and doesn't distinguish between the agent that requested a permission and the agent that ends up exercising it three hops later. CSA's framing treats this as the actual attack surface, not a theoretical one. The paper positions itself as an extension of CSA's earlier identity work, expanding scope toward a framework tailored specifically to agentic systems rather than adapting human-identity tooling after the fact. It pairs the IAM proposal with the MAESTRO threat-modeling framework for mitigation guidance, giving architects a way to map identity failures to specific layers of an agent stack rather than treating 'agent security' as one undifferentiated problem. What's notable is the timing relative to adoption. Enterprises are shipping multi-agent systems now, using identity infrastructure designed for human logins and simple service accounts, while the standards body writing the replacement is still in the blueprint stage. The 44 percent API-key figure suggests the gap between deployment and appropriate identity infrastructure isn't closing on its own — it's a design debt accumulating in production systems today, not a future risk to plan around later.