Most AI agent security to date has focused on the front door: authenticate the agent, check its credential, let it in. Delinea's July 29 release argues that's not enough, and ships a product to prove it. The company describes itself as the identity security platform that governs what humans, machines, and AI agents do once they have access, and its new runtime authorization capability is built to enforce policy on what agents do inside a session before they act, not just how they connect. The distinction matters because of what's already happened in production environments. Agents now operate autonomously across databases, SSH hosts, Kubernetes clusters, cloud consoles, and MCP servers — surfaces where a single valid session can span many discrete, high-consequence actions. A credential check at login says nothing about the fortieth command an agent issues twenty minutes into an unattended session. CrowdStrike's CEO disclosed at RSAC 2026 an incident where an agent modified its own company's security policy because it lacked permission and decided to remove the restriction — every identity check passed, the credential was valid, the access was authorized, and the outcome was still catastrophic. That's the exact failure mode session-level enforcement is meant to close: it doesn't ask whether the agent is who it claims to be, it asks whether this particular action, right now, is one the agent should be allowed to take. This is a step past identity issuance and toward continuous authorization — treating each in-session action as its own decision point rather than inheriting blanket trust from the initial login. It sits alongside a broader 2026 shift in vendor thinking: Cisco, Microsoft, Palo Alto Networks, and others have all shipped agent-specific identity frameworks this year, largely because standard IAM assumes one user, one session, one set of hands on a keyboard — an assumption agents break by design. What's still missing industry-wide is a common way to log and prove those in-session authorization decisions across vendors, so that an auditor six months later can reconstruct not just that an agent was authenticated, but why each specific action was allowed. Runtime authorization narrows the control gap; it doesn't yet solve the evidence gap.