JumpCloud proposes framework for managing AI agents

JumpCloud has introduced a new framework aimed at addressing the increasing challenge of managing and securing artificial intelligence (AI) agents within modern organizations. The company highlights that AI agents, which perform tasks such as Salesforce entries or infrastructure provisioning, are effectively part of the workforce but often lack traditional governance processes.

According to JumpCloud's research, non-human identities now outnumber human users in 83% of organizations. Only 21% of companies have specific governance controls in place for these agents. The new framework is divided into four stages: discovering agents, registering them as formal identities, managing access based on the principle of least privilege, and continuously governing their behavior.

The first stage focuses on identifying and documenting all agents operating within an environment, including their access privileges and functions. This is crucial for preventing the spread of "shadow AI," which refers to agents operating without formal records. Agents should then be registered as formal identities in a centralized directory, linked to a designated human owner responsible for their actions. This enables access control and the elimination of "zombie agents."

The third stage emphasizes the principle of least privilege, granting agents only the access necessary to perform their defined tasks. This includes employing time-bound credentials and regular review mechanisms. The final stage stresses continuous monitoring of agent behavior post-deployment, encompassing log data collection, periodic access reviews, and anomaly detection before issues escalate.

JumpCloud states that this four-stage model offers a solution to the rising risk associated with the increased use of AI agents and the lack of governance. Successful implementation requires unified identity management and security controls.