Delinea launches runtime authorisation for AI agents

Delinea has introduced runtime authorisation for AI agents, applying policy controls to agent actions before they are executed.

The addition extends Delinea's identity security platform into an area of AI use that many companies are still trying to manage: what software agents do after they gain access to systems such as databases, cloud consoles, Kubernetes clusters and SSH hosts.

Rather than focusing only on the point at which an agent connects, the new controls assess each action within a session. That means an organisation can allow, block or escalate individual commands or tool calls in real time across multiple environments.

The feature also covers systems connected through the Model Context Protocol, or MCP, alongside more established infrastructure access methods. Delinea argues that existing approaches tend either to validate sessions when they begin and review them later, or to observe only traffic that passes through their own servers.

Inside sessions

The release targets a growing operational issue for companies adopting AI agents in production settings. These agents can act autonomously across business systems, but their access patterns do not always resemble those of human users, and a single session can include many commands with different levels of risk.

Delinea's platform enforces policy on each action separately rather than treating the whole session as a single event. That can mean allowing one tool call, blocking another, or requiring human approval before a higher-risk action proceeds.

The vendor also injects credentials only at the moment of access and withdraws them when the task ends. In practice, the agent does not retain a standing credential between sessions, which Delinea says reduces the chance of an exposed credential being reused.

Another part of the launch is a distinction between human-driven and agent-driven connections. The platform identifies the type of connection before access is granted, then applies policy intended for that type of identity.

Audit trail

The system records actions at a detailed level, including tool calls, database queries and SSH commands, and ties them to a named identity. That record is intended to give security and operations teams a clearer audit trail when reviewing how an agent behaved inside production systems.

Art Gilliland, Chief Executive Officer at Delinea, framed the move as a response to a shift in security risk created by autonomous software.

"The security industry spent years solving credential theft, but AI agents have introduced a new problem: authorised access doing unauthorised things," said Art Gilliland, Chief Executive Officer, Delinea.

"You can have perfect credential hygiene and still have an agent tear through your production environment in milliseconds. Recording what happened or revoking access after the fact doesn't stop that - enforcing policy on the action as it runs does. The perimeter has moved to inside the session, and no one has figured out how to address that until now," Gilliland said.

The controls work through a single policy point across different protocols and connection types. That is meant to give security teams a common view of agent activity whether the agent reaches systems directly or through another service.

For customers, the issue is not only whether AI agents can connect securely, but whether staff can understand and account for every action once those agents begin operating across live environments. Delinea positions the new feature as a way to close that gap between secrets management, privileged access and operational oversight.

One customer cited production use as a reason for taking a closer look at action-level controls.

"AI agents were accessing our production systems and we had no way to know what they were actually doing once they were in," said Dhruv Kumar, Infrastructure Product Manager, Telnyx.

"With Delinea, every action is now tied to a named identity and recorded at the action level. That's what gave us the confidence to keep moving forward with AI without taking on risk we couldn't see," Kumar said.

The launch reflects a broader shift in identity and access management as AI agents move from experimentation into routine use in infrastructure and operations. In those settings, the challenge for suppliers is becoming less about authenticating a connection and more about controlling what happens next.

Delinea says the runtime authorisation feature is available as part of its existing platform and does not require separate infrastructure.