Identity and access management company JumpCloud Inc. today extended its Agentic IAM service to give information technology teams direct control over the autonomous artificial intelligence agents running inside their organizations.

Agents now sit in the same directory as employees and their laptops. Every agent is tied to a named human owner, and an unowned agent will not run at all. JumpCloud tracks each one from creation through to retirement.

Four capabilities make up the release. Every agent registered with JumpCloud gets a unique identity, which lets an administrator audit what it did or cut off its access without locking out the employee who deployed it. The platform will also hunt for Model Context Protocol servers set up outside official channels and route them through the JumpCloud AI Gateway.

Device trust and access policies that already govern sanctioned software now extend to AI tools. A privileged access management server built on the protocol rounds out the set. JumpCloud said that one will ship before the end of the fourth quarter. It’s meant to retire the shared application programming interface keys agents currently pass around.

Privileged access rules now apply to agents directly. Each grant expires with the task it was issued for. Nothing broader gets handed out in the meantime. That, JumpCloud said, is what lets companies point AI at legacy systems without rebuilding them.

Only 21% of organizations have put governance controls around nonhuman identities. That is the lowest score of 10 recommended AI security practices in JumpCloud’s quarterly survey of IT leaders, published in July. Nonhuman identities outnumber people at 83% of the organizations surveyed.

Joel Rennich, JumpCloud’s senior vice president of product management, said AI agents are “changing how identity management works.” The tools move fast, he said, and the information technology department has to know which agents exist, who owns them and what they can touch.

Agents do not simply reach into systems, according to Rahul Kamdar, senior vice president of product at JumpCloud. They “act inside them.” That, Kamdar said, makes identity the control plane for AI security.

Agentic IAM launched in April. JumpCloud launched a version on Google Cloud in June, expanding an existing partnership with the cloud provider. JumpCloud shipped its first AI features in January, shadow AI detection among them.

JumpCloud is based in Louisville, Colo. Investors have put $408 million into the business across 10 rounds. Sapphire Ventures led the most recent, a $159 million Series F, five years ago this month. The valuation then was $2.56 billion.

Image: JumpCloud