As artificial intelligence moves from assisting employees to performing tasks across enterprise applications, cloud infrastructure and development environments, organizations are facing a new cybersecurity challenge: determining what an AI agent is allowed to access, how that access is controlled, and whether its permissions remain appropriate as its actions change. The issue is gaining attention as AI agents increasingly operate with access to enterprise systems and resources, while security teams work to establish appropriate identity, access and monitoring controls. The Cloud Security Alliance has identified insecure identities and machine permissions as a leading cloud-security concern in 2026, noting that the growing number of non-human identities is expanding the potential attack surface.
Adarsh Reddy Bilipelli, a technology professional whose work spans DevSecOps, cloud security, data science and AI-driven cybersecurity research, brings experience relevant to these emerging challenges. His professional work includes secure CI/CD automation, cloud infrastructure, vulnerability remediation, monitoring and security controls, while his research examines the application of artificial intelligence and machine learning to cybersecurity problems.
Bilipelli's professional experience includes work with Star Technologys, an IT consulting and software development firm based in Fremont, California. The company offers a range of enterprise technology services, including staffing solutions, IT consulting, software development and cloud computing, working with clients on both specific projects and longer-term engagements.
The growing use of AI agents is changing the traditional understanding of identity inside enterprise systems. For years, identity and access management primarily focused on employees, contractors, applications and service accounts. AI agents introduce another category: software systems capable of making decisions, selecting tools, calling APIs and carrying out multi-step tasks with limited human intervention.
That change creates a practical security question. An organization may know which employee or application initiated a process, but it can be more difficult to determine which actions were performed by an autonomous agent, what permissions it used and whether those permissions were necessary for the task.
Recent Cloud Security Alliance research has highlighted these concerns. Its March 2026 study found that more than two-thirds of organizations could not clearly distinguish AI-agent activity from human activity, while the research also identified widespread concerns around excessive permissions and new access pathways created by AI agents.
For Bilipelli, this raises the need to treat AI-agent access as part of the broader security architecture rather than as an isolated AI issue.
“AI agents should not be treated simply as another application running inside the enterprise. When an agent can make decisions and interact with multiple systems, security teams need to understand not only what identity the agent uses, but what permissions it has, why those permissions are required and how those permissions are being used,” Bilipelli said.
The principle of least privilege becomes particularly important in this environment. AI agents should receive only the access required to perform a defined task, with permissions reviewed as their responsibilities change. Cloud Security Alliance research has pointed toward more specific identity controls for AI agents, including lifecycle management, appropriate assurance levels and controls designed around the way autonomous systems operate.
This challenge becomes more complicated as organizations connect AI agents to cloud services, APIs, databases, development tools and internal business applications. A single agent may interact with several systems during one workflow. If credentials are over-permissioned or poorly monitored, an error, compromised credential or manipulated instruction could potentially extend beyond the agent's original purpose.
Bilipelli's professional experience in DevSecOps is relevant to this challenge because his work has involved integrating security practices into software delivery and infrastructure processes. His profile includes experience with CI/CD pipeline automation, security scanning, vulnerability review, deployment controls and cloud infrastructure reliability.
He has also emphasized the importance of moving security earlier into the engineering lifecycle. Rather than waiting until an application is ready for production, security checks can be incorporated into repositories, build pipelines, infrastructure configurations and deployment workflows. This approach can help organizations identify vulnerabilities and configuration issues earlier while maintaining repeatable development and deployment processes.
Monitoring is another important part of the equation. When software begins acting with greater autonomy, organizations need visibility into what those systems are doing. Logging and observability can help security and engineering teams identify unusual activity, failed deployments, infrastructure problems and other deviations from expected behavior.
Bilipelli's experience includes cloud observability using tools such as Azure Monitor, Log Analytics and Application Insights. His work in this area has focused on improving visibility and helping teams detect abnormal behavior, system degradation and infrastructure issues earlier.
“Identity, access and monitoring cannot be treated as separate controls when software is increasingly capable of acting autonomously,” Bilipelli said. “Organizations need to understand the context of an agent's actions and establish controls that can limit unnecessary access while providing enough visibility to investigate unexpected behavior.”
The industry response is beginning to move beyond simply identifying AI agents toward controlling what they can do. Emerging security guidance recommends giving autonomous agents distinct identities, limiting permissions to the minimum required for a task, using short-lived credentials and applying stronger isolation between agents and the systems they can access. These controls are intended to reduce the consequences of a compromised credential, an unexpected action or an agent operating outside its intended scope.
Cloud providers and technology organizations are also developing security mechanisms specifically for agentic workloads, including identity-aware access controls, network-level boundaries, runtime protections and policy enforcement. The shift reflects a broader recognition that conventional application security controls may not always be sufficient when software can independently interpret instructions, select tools and execute actions across multiple systems.
For organizations, the practical challenge will be finding a balance between autonomy and control. Restricting every AI agent so heavily that it cannot perform useful tasks could undermine the value of automation, while granting broad and persistent permissions can increase security exposure. The emerging approach is therefore focused on making access more contextual, traceable and temporary, with monitoring and policy controls remaining active throughout an agent's operation.
Bilipelli plans to continue exploring the intersection of cloud security, DevSecOps and artificial intelligence, with a focus on how security practices can adapt as autonomous systems become more integrated into enterprise infrastructure. His broader work combines secure software delivery, cloud reliability, vulnerability management, data science and AI-based cybersecurity research.
As AI agents become more capable, the question for enterprise security will increasingly shift from whether organizations should use them to how safely they can operate them at scale. Protecting identities, limiting permissions, maintaining visibility and establishing clear operational boundaries will be essential to ensuring that greater automation does not create greater exposure.
