agentic – Tutor Dasher

In the current state, identity is predominantly focused on humans. Traditional identity and access management (IAM) systems were developed for a world where human users and static applications were the norm. Identities were managed using models like role-based access control (RBAC) and multifactor authentication (MFA), with decisions made at the time of login. Even with the move towards zero-trust, the underlying assumption remains the same: identities are well-defined, limited, and relatively stable. However, the emergence of agentic AI systems challenges these assumptions. The shift to agentic systems has fundamentally changed the security landscape. We are now not only safeguarding "users"; we are securing a vast, autonomous network of non-human identities (NHIs) that operate at machine speed. Autonomous agents engage with tools, access APIs, create sub-agents, and function across multiple domains without direct human involvement. These agents often rely on shared credentials, short-lived tokens, or implicit trust boundaries, leading to identity ambiguity, weak attribution, and increased attack surfaces. In essence, the current IAM framework is not aligned with the dynamic, autonomous nature of AI agents. The need for a new identity stack The advent of agentic AI systems brings about a new category of identities - autonomous, non-human entities such as AI agents, bots, and services that operate independently, dynamically, and at scale. Unlike human identities, these entities can be created on-demand, assign tasks to other agents, and interact across various systems without direct supervision, posing challenges for attribution, control, and trust. For instance, agents operate faster than human oversight, and the 'kill switch' has evolved from a button to an autonomous circuit breaker. Traditional identity models, designed around static users and roles, are inadequate to govern this fluid ecosystem. Therefore, there is a critical necessity for an evolved identity framework that can uniquely identify these entities, trace their origins, enforce precise and contextual access controls, and continuously verify their actions to ensure secure and responsible operations. A look into the modern identity stack for agentic systems Agent identity and provenance: Each AI agent must possess a distinct, verifiable identity linked to its source, whether created by a human, system, or another agent. Provenance enables traceability, allowing organizations to determine who initiated an action and under what authority. This establishes accountability and prevents anonymous or rogue agent behavior. Ephemeral credentialing: Instead of long-lasting credentials, agents should utilize short-lived, task-specific tokens that are automatically generated and revoked. This reduces exposure in the event of a breach and aligns access strictly with the duration and scope of a task. It enforces the zero-standing privilege (ZSP) principle. Contextual Authorization: Access decisions should be dynamic and based on real-time context, such as behavior, environment, and risk indicators. Rather than fixed roles, permissions should adapt continuously to the actions and location of the agent, ensuring tighter and more relevant controls. Delegation and chain of trust: Agentic systems often involve multiple levels of delegation encompassing user communication with agents and agent communication with tools. A clear and enforceable chain of trust is essential to track authority and restrict the spread of permissions, thus preventing privilege escalation. Identity threat detection and response (ITDR): Systems must constantly monitor agent activities, reassess risks, and adjust permissions in real-time. For example, continuous verification now detects semantic drift, where an agent's actions gradually deviate from its original intent or authorized purpose. This helps identify subtle misuse, compromised workflows, or manipulated prompts that may not trigger traditional security alerts. Observability and attribution: A robust audit trail is crucial for capturing who executed which action, through which agent, and with which tools. This level of visibility ensures accountability, facilitates incident response, and instills confidence in autonomous systems by making their actions transparent and comprehensible. Identity as a real-time control plane in agentic systems Identity is set to evolve into a real-time control mechanism for agentic systems, beyond just a gateway for access. Key transformations will include: Identity becomes behavioral as trust is continuously evaluated rather than statically assigned. Agents are treated as primary entities, managed, governed, and audited akin to human users. Policies need to be adaptive as AI-driven policies evolve alongside threats and usage patterns. Zero-trust evolves into zero-standing privilege, where access is granted only for the duration of a verified task. Identity integrates with execution frameworks to authenticate, authorize, and log every tool interaction. Inference The rise of agentic AI systems necessitates a fundamental reevaluation of identity management. Static credentials and perimeter-based trust mechanisms are no longer adequate. Agent identity management requires a shift from RBAC to ABAC. The new identity stack must be dynamic, contextual, and deeply integrated into the operational fabric of AI systems, ensuring that every action, whether initiated by a human or an autonomous agent, is verifiable, traceable, and inherently secure.

Tag: agentic

A new frontier: Identity stack evolves for agentic systems

Splunk.conf: Cisco and Splunk expand agentic SOC vision

Salesforce pays $8bn for Informatica to underpin agentic AI play

Few UK organisations ready to seize agentic AI opportunity

Quick Links

Resources

Support