Governing Agentic AI through Identity and Access Management Capabilities and Principles
Majander, Mikko (2026)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202604298482
https://urn.fi/URN:NBN:fi:amk-202604298482
Tiivistelmä
The increasing adoption rate of agentic AI systems and tools in enterprise environments poses possible security and compliance risks if not managed properly. Several layers of safety mechanisms can be introduced, such as prompt templates or design decisions on tool usage for safe implementation. To further reduce the access risk to a acceptable level, it is necessary to consider including agentic AI into the Identity and Access Management (IAM) governance framework and design what IAM controls can be used to manage the lifecycle, authentication and authorization of agentic AI entities alongside other human and non-human identities.
Literature review was done to research and understand the current protocols and technologies enabling the agentic AI tool usage and different authentication and authorization scenarios. Most complex scenarios include agents discovering agents and delegating their workload. From these kind of scenarios it was possible to see that oversight is challenge due to the volume of activities and ephemeral nature of agentic systems: one task of action could be performed by multiple agents in non-deterministic way.
The work evaluated, how well IAM-principles, such as governance, and technical capabilities for authentica-tion and authorization can be used to reduce the access risk.
The outcome was that IAM governance model and framework should have the agentic AI solutions in the scope and the ownerships and responsibilities should be possible to model to govern the solutions. Likewise, the conventional technical capabilities can be mostly utilized. Due to the short lifespan and potentially non-deterministic tool and information usage, there is emphasis on leveraging the runtime access management capabilities to make the authorization decisions and tools to support elicitation to keep confidential information away from agents.
Literature review was done to research and understand the current protocols and technologies enabling the agentic AI tool usage and different authentication and authorization scenarios. Most complex scenarios include agents discovering agents and delegating their workload. From these kind of scenarios it was possible to see that oversight is challenge due to the volume of activities and ephemeral nature of agentic systems: one task of action could be performed by multiple agents in non-deterministic way.
The work evaluated, how well IAM-principles, such as governance, and technical capabilities for authentica-tion and authorization can be used to reduce the access risk.
The outcome was that IAM governance model and framework should have the agentic AI solutions in the scope and the ownerships and responsibilities should be possible to model to govern the solutions. Likewise, the conventional technical capabilities can be mostly utilized. Due to the short lifespan and potentially non-deterministic tool and information usage, there is emphasis on leveraging the runtime access management capabilities to make the authorization decisions and tools to support elicitation to keep confidential information away from agents.