How Microsoft Entra aims to keep your AI agents from running wild

IT Professional Comments Off on How Microsoft Entra aims to keep your AI agents from running wild

The architectural shift here is not about adding another AI dashboard. It is about treating an agent as a first-class identity object in the same control plane as people, apps, and devices. For teams that already rely on Entra-style IAM, the practical win is consistency: one governance model for authentication, authorization, and lifecycle management across human and non-human actors.

That consistency also creates a deployment dependency. Agent identity only becomes useful if agent creation, registration, and revocation are wired into the systems where agents are actually born. In Microsoft-centric environments, that means development platforms, security tooling, and endpoint controls need to feed the same registry. In hybrid estates, the harder problem is not the identity record itself but the cross-system traceability needed to know which agent touched which resource, and under what authority.

The security value is strongest where least privilege can be enforced automatically. Agent behavior is often dynamic, so static role assignment may be too blunt once agents start chaining tasks or spanning multiple datasets. That raises an operational trade-off: tighter controls improve containment, but they can also slow automation unless policy is engineered around task scope, not just tenant membership.

For IT leaders, the key question is governance maturity. If shadow agents can appear outside sanctioned workflows, the registry becomes only part of the answer. Organizations will need clear onboarding paths, API-based inventory reconciliation, and incident response rules for anomalous agent activity. Otherwise, identity management becomes descriptive rather than preventive, which limits its value in regulated or high-risk environments.

The array of AI-related announcements that came out of Microsoft’s Ignite Conference was so dizzying that it was too easy to miss the significance of certain launches that weren’t as sexy as others. Buried in that tidal wave was news of something called Entra Agent ID, the main idea of which is to use Microsoft Entra to govern AI agents in the same way that Entra currently governs human users; that is, to give each agent a unique, managed identity and apply familiar Entra identity controls such as conditional access, identity governance, and identity protection. Entra is Microsoft’s cloud-based identity access management (IAM) solution. Also: How Microsoft’s new security agents help businesses stay a step ahead of AI-enabled hackers This idea of “personhood” equivalence for AI agents, as my colleague David Gerwitz described it (see Microsoft’s new AI agents won’t just help us code, now they’ll decide what to code), is also getting some airplay from the OpenID Foundation as well as Original Postress-releases/okta-introduces-cross-app-access-to-help-secure-ai-agents-in-the/" target="_blank" rel="noopener nofollow" shape="rect">from Okta, a Microsoft IAM competitor. In the same way that IAM systems like Microsoft’s Entra have been traditionally used to provision human users with digital identities and access to business resources, there’s a growing belief that those same IAM systems should be used to manage the access that AI agents are afforded to those same organizational systems. Although organizational AI agent deployment is currently in a nascent state, the urgent need to consider such an identity-centric approach is brought about by an expected behind-the-firewall proliferation of both sanctioned AI agents as well as their unsanctioned shadow IT counterparts.

Agents, everywhere

Today, the number of users greatly outnumbers the number of currently active agents. However, as business-oriented agent development and deployment becomes relative child’s play through tools such as tasklet.ai, even average users seeking modest productivity gains will, in true shadow IT style, be inclined to put such agents to work on their behalf. According to IT research firm Gartner, 42% of respondents to its 2026 CIO and Technology Executive Survey said that their enterprises plan to deploy AI agents within the next 12 months. A Gartner spokesperson told ZDNET that by 2030, CIOs expect that 0% of IT work will be done by humans without AI, 75% will be done by humans augmented with AI, and 25% will be done by AI alone. Also: Ignite 2024 introduces new AI agents and more for Microsoft 365 Copilot Between that and executive pressures to harness all that AI has to offer and gain a competitive edge, the ratio of users to agents could easily flip to the point that agents (some of which will operate with a fair amount of autonomy) could outnumber human users by several orders of magnitude. Whereas human users come and go and IAM systems are finally mature enough to keep up with both hiring and attrition (relying on open standards like the System for Cross-domain Identity Management aka “SCIM” to bridge the gap between HRMS and IAM systems), the ephemerality of AI agents — some of which may last no more than a few seconds — will also challenge traditional norms of ID management and access control. To help organizations get a jump on agent proliferation before they fell too far behind, Microsoft first previewed Agent ID in May of this year at its Build conference. But Microsoft corporate vice president of AI Innovations Alex Simons told ZDNET that it was basically a toy at that point — little more than an agent tagging scheme.

Enter Entra

Now, six months later at Ignite, Entra Agent ID has evolved into a full-blown agent identity management layer within Microsoft’s larger Agent 365 AI control plane that cuts across Microsoft’s ecosystem of AI-infused platforms. As shown in the screenshot below, the Agent ID dashboard is now available through Microsoft Entra’s left-hand navigation.
Within Microsoft’s cloud-based Entra ID identity provisioning and management system (traditionally used for managing the intersection of users and the resources they need access to), “Agent ID” appears as an option in the left-hand navigation. When selected, a top level dashboard that summarizes all known AI agents appears and from there, IT administrators can drill down on the basis of agent category, active status, creation data and other criteria.
Source: Microsoft For example, in other parts of that ecosystem where Microsoft platforms such as Copilot Studio and Azure AI Foundry are used to develop and deploy AI agents, those agents are automatically and uniquely registered in the Entra Agent Registry. The same is true of other AI agents that enter the organization through other parts of Microsoft’s AI fabric. For example, also at Ignite, Microsoft announced the availability of a slew of pre-built security agents — some from Microsoft and others from partners — to help businesses stay a step ahead of AI-enabled hackers. As customers choose to enable those agents through a series of storefronts that contextually make their appearances with Microsoft Entra, Purview, Defender, and Intune, their presences on the corporate network will also be automatically registered with Entra Agent Registry. For agents built or offered outside of the Microsoft ecosystem, the Entra Agent Registry will be updatable via Microsoft’s RESTful Graph API whose list of available methods has been expanded specifically for interfacing with the registry (the feature is currently in beta). However, as for other AI agents that employees choose to enlist outside of IT’s guardrails, there is currently no manual way (eg, a form) to update the registry. Also: Enterprises are not prepared for a world of malicious AI agents Presumably, however, if the resources in question (the ones that an agent needs access to) are already subject to some form of central oversight, certain controls may already be in place that prevent access by anything other than authorized identities (humans, agents, or otherwise). “We’ve extended [Entra] to manage agents, and it really solves three sets of challenges for customers,” Simons told ZDNET. “First, is just getting a handle on where the heck are all of my agents. Which ones are they and what are they capable of doing? Second is to get a unique identifier for each of those agents so you can see what it is doing across your whole estate. For example, if it’s trying to gain access to SharePoint documents or some data in Azure or AWS. And third is to manage the permissions of those agents and making sure that they can that they have a least privilege model where those agents are only allowed to do the things that they should do. If they start to do things that are weird or unusual, their access is automatically cut off.” In the bigger picture, Microsoft sees the capabilities of Entra Agent ID and the wider Agent 365 control plane as a big deal where compliance matters. For example, keeping a lid on data access in highly regulated environments where an agent that runs amok could result in serious legal consequences (never mind the possibility of exposing sensitive data and the potential for irreparable harm to downstream victims). In addition to inheriting agent-specific functionality from Microsoft Entra as described above, the Agent 365 control plane also inherits other agent-specific capabilities from Defender, Intune, and Purview. Also: 3 ways AI agents will make your job unrecognizable in the next few years The public preview that was demonstrated at Ignite is available to Microsoft’s customers at no additional cost, Simons said. However, once they are made generally available on a commercial basis in the first quarter of 2026, usage will involve incremental fees. Microsoft Entra itself is licensed on a per-user per-month basis. For Agent ID, Microsoft is currently exploring several ideas with customers, one of which is based on an agent’s volume of activity (the degree to which the agent acts on other organizational resources on behalf of one or even more users). But the exact business model and pricing haven’t been announced yet.
https://www.zdnet.com/article/how-microsoft-entra-aims-to-keep-your-ai-agents-from-running-wild/

Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.