Governing AI Agents Starts With the Infrastructure Beneath Them

AI agents are moving into infrastructure operations. They provision compute and storage, manage backup jobs, update configurations and respond to incidents at machine speed. Each of these actions touches production systems, and regulated organizations need the same confidence in an agent's actions that they expect from any other change to a critical environment.

The Gartner® report, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, is available for complimentary access. According to Gartner:

“An agent that makes a wrong API call does not produce a wrong answer in a chat window. It may delete a storage volume, misconfigure a firewall rule, exhaust a cloud budget, or trigger a failover cascade.”

Source: Gartner, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, Chandra Mukhyala, 16 July 2026.

Should organizations block AI agents or govern them?

Some organizations respond to agent risk by restricting AI entirely. Teams still face pressure to automate, however, and unsanctioned agents can appear in production with little oversight. According to Gartner:

“The right posture is a controlled yes: I&O establishes the guardrails, defines the authorized scope, and approves agents for deployment within those bounds.”

Source: Gartner, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, Chandra Mukhyala, 16 July 2026.

Rackspace recommends that every agent that touches production infrastructure have five things before it goes live:

• A named owner accountable for its outcomes and security
• Its own identity, with credentials that expire
• Permissions limited to the task it performs
• Enforced policies, with human approval for high-risk actions
• A tested way to halt it that does not depend on the agent itself

Three layers of agent control

Source: Gartner, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, Chandra Mukhyala, 16 July 2026.

This graphic was published by Gartner, Inc. as part of a larger research document and should be evaluated in the context of the entire document. The Gartner document is available upon request from Rackspace Technology.

Where does agent governance live?

Many of these controls depend on the infrastructure beneath the agent. Identity services issue and revoke agent credentials, network isolation limits which systems an agent can reach and separate identities for production data and backups keep a single agent from affecting both. Observability data from agent activity feeds the security information and event management (SIEM) and IT service management (ITSM) tools teams already use, and cost controls set limits on how much infrastructure an agent can consume.

Agents also depend on accurate context. When an agent calculates the impact of an action from outdated configuration data, its decisions inherit that error. Governed, current data foundations give agents the context they need to act within intended boundaries.

What happens when an AI model changes?

AI models can change or become unavailable with little notice. Organizations can reduce that exposure by running open models on private infrastructure they govern, customizing those models with their own data and using an orchestration layer that allows them to switch models while preserving context.

Agent governance is operational resilience

An AI agent that exhausts a cloud quota or triggers an unintended failover can disrupt critical services. Controls that halt, contain or limit agent actions belong in operational resilience planning alongside recovery testing and incident response, with regular testing to verify their effectiveness.

How Rackspace helps

Rackspace Technology operates infrastructure and AI as one governed system in private and hybrid environments. Dedicated GPU infrastructure, managed private inference and model customization on the customer's own data keep models, prompts and outputs inside the customer's boundary. Model-agnostic orchestration supports continuity when models change, and agents run with identity integration, role-based access controls and observability connected to existing tools.

Rackspace Forward Deployed Engineers embed inside the customer's environment to put these controls into production. They establish agent identity standards, guardrails, logging and containment drills as part of the systems they deliver, and they remain accountable after go-live. The customer owns the environment, the data and the control plane, while Rackspace operates the stack 24x7x365.

Questions leaders ask

How should organizations classify the risk of AI agents?

“AI agents should be treated as a distinct category of operational risk, with the same rigor applied to human error, hardware failure, or cyberattack.”

Source: Gartner, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, Chandra Mukhyala, 16 July 2026.

Our perspective. Rackspace places agent controls inside the same operating model as the rest of the stack, with identity integration, role-based access, observability connected to existing tools and recovery that has been tested.

Is agent governance a compliance cost?

“I&O should frame governance infrastructure in FinOps terms, primarily as a cost-control mechanism preventing runaway AI-driven infrastructure spend, not as compliance overhead.”

Source: Gartner, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, Chandra Mukhyala, 16 July 2026.

Our perspective. Dedicated, right-sized GPU capacity and managed AI operations give organizations predictable infrastructure economics alongside the controls that govern how agents use that capacity. 

Which controls stop an agent that goes wrong?

“Kill switches, circuit breakers, and blast radius containment are the operational resilience tools for AI-driven I&O infrastructure.”

Source: Gartner, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, Chandra Mukhyala, 16 July 2026.

Our perspective. Rackspace Forward Deployed Engineers build containment, logging and halt procedures into the systems they deliver and help rehearse them alongside recovery testing. 

Read the research

Get complimentary access to the Gartner report, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations. Then read the Rackspace white paper, Sovereignty and governance across the full stack, to see how governed AI fits with sovereignty, resilience and private and hybrid infrastructure.

Gartner attribution and disclaimer

Gartner, AI Agent Management Control Plane: Evaluation Guide for Infrastructure and Operations, Chandra Mukhyala, 16 July 2026. 

GARTNER is a trademark of Gartner, Inc. and/or its affiliates. Graphics in this document were published by Gartner, Inc. as part of larger research documents and should be evaluated in the context of the entire document. The Gartner documents are available upon request from Rackspace Technology. Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner's research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.