Chapter 3
Operating and monitoring AI agents: Identity-first access
Securing agentic AI requires governance across its entire life cycle, beginning with identity.
AI agents operate as digital identities that authenticate, access systems, and execute actions across environments. Every action depends on credentials and permissions, making identity the foundation of AI security.
Unlike scripted automation, agentic AI dynamically accesses systems and invokes tools across workflows, relying on identities and credentials to authenticate and act.
Machine identities already outnumber human users in most environments. Ponemon research shows that 51% of organizations consider managing nonhuman identities a high priority as automation expands.4 Service accounts, APIs, automation scripts, containers, and application credentials continue to proliferate. AI agents accelerate this growth, often without centralized oversight or life cycle governance.
4 Ponemon Institute, Managing Risks and Optimizing the Value of AI, GenAI & Agentic AI, March 2026.
of organizations consider managing nonhuman identities a high priority.
The risk of identity sprawl
Research on nonhuman identities shows that organizations are struggling to manage this expansion. Machine credentials often remain active long after their purpose ends, privileges accumulate over time, and secrets are stored in insecure locations. Each unmanaged identity expands the potential attack surface.
AI agents amplify these challenges, because they operate dynamically. They may authenticate across multiple systems, access sensitive data, and initiate actions that influence business outcomes. When identity governance lags behind deployment, the resulting visibility gaps can obscure risk.
Managing nonhuman identities has become a priority as organizations confront excessive privileges, poor life cycle management, and exposed credentials. In environments where a single human identity may be associated with dozens of machine identities and service tokens, maintaining control becomes significantly more complex.
Typical identity risks include overprivileged service accounts, exposed credentials, unmanaged API keys, orphaned identities, and poor audit visibility. These conditions create opportunities for attackers to exploit credentials rather than breach systems directly.
5 https://www.cybersecuritytribe.com/news/research-reveals-44-growth-in-nhis-from-2024-to-2025
Enforcing least privilege for autonomous systems
For example, an AI agent with overly broad access to financial systems could expose sensitive data if its credentials are compromised. What appears to be an efficiency gain can quickly become a security liability.
Attackers increasingly target credentials, because they provide direct access to trusted environments. When an AI agent operates with elevated permissions, compromised credentials can allow adversaries to move laterally across systems while appearing legitimate.
Strong identity governance provides the foundation for control and accountability. Assigning unique identities to agents ensures traceability and auditability. Limiting permissions to defined tasks reduces exposure while preserving operational flexibility. Context-aware access policies help ensure that agents operate only within approved boundaries.
Identity-first access also supports zero-trust principles. Rather than assuming trust based on network location or system role, access decisions are evaluated based on identity, context, and behavior. This approach reduces reliance on perimeter defenses and strengthens resilience against credential misuse.
An AI agent that provisions accounts for new hires illustrates the importance of life cycle controls. The agent may require temporary access to human resources systems, directories, and collaboration platforms. Without defined expiration policies, elevated permissions can persist long after onboarding tasks are complete, creating hidden risk over time.
Life cycle governance ensures that identities are created, monitored, and retired according to policy. This includes tracking credential issuance, enforcing rotation schedules, and removing access when roles or workflows change. Automation can support these processes, but oversight remains essential.
With more autonomous systems in production, identity governance must extend beyond authentication. It must also encompass visibility, accountability, and continuous oversight.
Issuing unique identities for each agent and automation process
Enforcing least-privilege permissions aligned with defined tasks
Implementing credential rotation and secret management policies
Monitoring identity usage patterns for anomalies
Retiring unused identities and access paths promptly
Identity-first access establishes accountability across autonomous systems. It enables organizations to enforce authority boundaries and detect anomalies quickly. When identity is governed effectively, security teams gain the context needed to distinguish legitimate activity from misuse.
Enterprises should treat every AI agent as a governed digital identity. OpenText recommends issuing each agent a unique identity, enforcing least-privilege access, and applying strong credential and life cycle management controls from the start.
As AI agents authenticate across systems, access sensitive data, and take action dynamically, identity governance becomes essential to reducing risk. Organizations should secure and rotate credentials, retire unused identities promptly, and continuously monitor agent activity for anomalies or misuse.
With an identity-first approach, enterprises can improve accountability, reduce credential exposure, and scale agentic AI securely and with confidence.
Operate AI securely with OpenText identity and access management solutions.
