AI Agents Stealthily Emerge as Leading Users in Companies, Leaving Organisations Unprepared

Brief Overview

• AI agents are increasingly being adopted in organizations, frequently lacking appropriate governance.
• These agents manage sensitive information and demand rigorous access controls.
• Machine identities surpass human identities, complicating security management.
• Organizations face security risks from AI agents with excessive permissions.
• New standards such as MCP heighten the number of connections requiring secure oversight.
• Contemporary identity security must progress to encompass AI agents.
• Palo Alto Networks’ Idira aids in managing access across various identity types.

The Growth of AI Agents in Organizations

AI agents are swiftly becoming a fundamental part of the enterprise identity framework. However, numerous organizations are implementing them more quickly than they can regulate, leading to significant blind spots regarding the access capabilities of these agents, the systems they can connect to, and how that access is managed.

As AI agents increasingly engage with sensitive information and automate operational workflows on behalf of users, they are being integrated into customer service, software development, and knowledge management. The primary security concern is not just what these agents are capable of, but whether organizations truly grasp and govern the privileges they have been assigned.

A New Category of Privileged Identity

Many organizations still wrongly interpret AI agents as simple software applications. In reality, they function as digital workers, capable of retrieving data, initiating workflows, and interfacing with critical business applications. Whether linked to internal knowledge repositories or customer-centric platforms, these agents frequently handle extremely sensitive data. Nevertheless, distinct ownership and visibility of how these permissions are utilized remain significantly inadequate.

This issue arises at a moment when the overall identity landscape already faces considerable pressure. Palo Alto Networks’ 2026 Identity Security Landscape Report revealed that machine identities now exceed human identities by a ratio of 109 to 1 in Australia, with both machine and AI agent identities anticipated to surge in the upcoming year. As organizations expand their AI initiatives, they are adding a vast, intricate layer of non-human users that necessitate the same visibility, governance, and accountability that have traditionally been applied to human employees.

The Broadening Attack Surface

AI agents are not intrinsically risky; the danger arises when organizations assign enterprise-level privileges before implementing enterprise-level controls. Our research shows that Australian organizations estimate around 40% of AI agents and 41% of machine identities already have access to organizational data, including vital business systems.

An AI agent with excessive permissions presents an attractive target for cybercriminals. If breached, manipulated, or misused, it turns into a trusted entryway into the organization, enabling malicious actors to navigate laterally through networks, remove data, or interrupt essential operations.

This issue is compounded by emerging standards like the Model Context Protocol (MCP). While MCP is beneficial for scaling agentic AI by establishing a consistent method for agents to communicate with databases and enterprise applications, it concurrently amplifies the number of credentials, permissions, and trust relations that are active throughout the network.

Each new link between an AI agent and a business system generates an access route that must be protected. As organizations develop these interconnected AI ecosystems, insight into who or what possesses access to critical systems becomes as critical as securing the AI models themselves.

Advancing Governance for the Agentic Age

To reduce these risks, contemporary identity security principles must be broadened to encompass all identity types, including human, machine, and agentic.

This necessitates a thorough strategy: identifying and cataloging all AI identities, implementing lifecycle management, enforcing stringent least-privilege access, and consistently monitoring for unusual activity. It additionally involves ensuring human oversight for compliance and guaranteeing that access can be immediately revoked if an agent is compromised or decommissioned.

As AI agents become deeply integrated into everyday operations, traditional identity tools will become insufficient. Organizations need advanced identity security platforms that extend governance beyond typical human users. The industry is experiencing a profound transformation: transitioning from managing only human access to governing every single identity operating within the enterprise from a centralized control framework.

To assist organizations in navigating this transition, Palo Alto Networks has recently introduced Idira. By merging privileged access management with capabilities for machine and agentic identity security, Idira offers a unified platform to discover, secure, and govern access across all identity types.

AI agents are swiftly becoming the most interconnected and privileged entities within the enterprise. As Australian organizations continue to expand agentic AI, strong governance will shift from being an obstacle to becoming the ultimate facilitator of security, trust, and innovation.

Conclusion

As AI agents become increasingly integrated into enterprise practices, they introduce new challenges for identity management and security. Organizations must tackle the governance of these agents to avert security incidents and maintain compliance. The emergence of AI agents requires a transformation in identity security strategies, with platforms like Palo Alto Networks’ Idira offering the essential tools for managing an array of identity types.