Runtime authorization becomes the next layer of AI agent security

Identity security vendors are rapidly expanding beyond authenticating AI agents to governing what they can do after they gain access. New announcements from Delinea and startup Neo highlight growing industry interest in runtime authorization, where tool calls, API requests and privileged actions are evaluated before execution.

Delinea extends privileged access to AI agent actions

Delinea is adding runtime authorization for AI agents to its identity security platform. This extends policy control beyond initial authentication and access to the individual actions an agent attempts during a session.

The company says its platform can analyze tool calls, database queries, SSH commands, and interactions with Kubernetes clusters and cloud environments before the agent executes a specific action. Depending on the enterprise policy, the action can be allowed, blocked, or referred for human approval.

Runtime authorization addresses a gap left by traditional identity systems. Authenticating an AI agent and granting it access to a system does not mean it should be allowed to perform every action available once connected.

For example, an agent may need access to a production database to retrieve information, but that does not mean it should be able to delete records, modify permissions, or export unrestricted data.

“The security industry spent years solving credential theft, but AI agents have introduced a new problem: authorized access doing unauthorized things,” says Art Gilliland, chief executive officer at Delinea.

Delinea inserts a second authorization layer after authentication. Once an AI agent has established its identity and connected to a resource, the platform evaluates each requested action against enterprise policy before allowing it to proceed.

The company is combining these runtime controls with its existing privileged access capabilities. The solution can inject a credential “just-in-time” when access is required, scope it to the task, and revoke it when the task is completed.

The launch builds on Delinea’s acquisition of StrongDM earlier this year, bringing StrongDM’s runtime authorization capabilities into Delinea’s privileged access management platform.

Delinea is not alone. Nuggets recently introduced an Authority Control Plane that evaluates an agent’s identity, delegated authority, organizational policy and runtime context before approving actions, highlighting growing industry interest in action-level governance for enterprise AI agents.

Neo debuts with $100M to tackle agentic software security

Neo has emerged from stealth with $100 million in funding from Andreessen Horowitz and Bessemer Venture Partners, with participation from Craft Ventures and Merlin Ventures.

The company describes its approach as Agentic Software Control, which aims to help security teams manage AI agents and software gaining autonomous capabilities across the enterprise.

Neo approaches the problem from software visibility rather than privileged access management. The company says organizations first need to understand where autonomous capabilities exist before they can govern them.

Neo starts from a broader visibility problem, as agentic capabilities can appear in standalone AI agents, browsers, developer tools, SaaS applications, plugins, extensions, MCP servers, and conventional enterprise software.

“Organizations need visibility into what this software can do and the ability to govern its behavior in real time,” says Zane Lackey, general partner at Andreessen Horowitz.

The Neo platform identifies software running across the enterprise and analyzes what each application can do, what systems or data it can access, and the risks those capabilities may create.

It also provides attribution to associate activity with the human, agent, application, or identity. Organizations can then apply policies to tool calls, API access, data movement, and agentic workflows.

“Neo gives enterprises the real-time control layer they need to understand what agentic software can do, govern how it behaves, and secure adoption without slowing down the business,” says Nick Warner, chief executive officer and co-founder at Neo.

As organizations deploy autonomous software capable of making decisions and interacting with critical systems, runtime authorization is emerging as a foundational layer of AI security.

Article Topics

AI agents  |  Delinea  |  identity acceptance  |  identity control plane  |  identity security  |  NEO