AGENTIC AI FOR ENTERPRISE CLOUD OPERATIONS: AUTOMATION, GOVERNANCE, AND SECURITY RISKS

Abstract

Cloud computing has ushered in a radical shift in enterprise IT delivery to facilitate the use of IT resources. Moreover, cloud environments afford operational activities, which take care of provisioning, scaling, and management of IT resources, provided that user-defined policies are in place. The emergence of Artificial Intelligence technologies allows these management activities to be fully automated, which means that these activities, which are responsible for resource management and operation and mainly triggered by event notifications sent by the infrastructure provider, can now be executed without direct human supervision, enabling a higher degree of autonomy. Activities managed in this fashion can be considered agentic as in addition to performing tasks they can take decisions and act in a self-directed manner, as long as they meet a system-chosen goal.

The conditions in which these IT environments can be managed with a higher degree of autonomy must consider the implementation of a governance framework that establishes the authorization model for the agentic activities. Several roles are defined within the governance framework including functions that specify the policies on which the offence is constructed, integration with security services, and audit. The agentic capabilities introduced in the cloud environments also bring to the surface security concerns that demand extra control. Considering the threat landscape, agentic cloud environments are particularly vulnerable to Jacking, DIverting and MeTro (JIM) attacks, where the perpetrator takes control of these activities, diverts their execution, or transfers control to a rogue cloud providing by exploiting Identity & Access Management (IAM) keys or other management information associated with the cloud provider.