Commvault has announced a new capability to discover, protect and recover local configurations and data tied to generative AI tools and AI agents that employees use on Windows and macOS computers. The offering covers Claude Cowork, Claude Code and OpenClaw, and aims to prevent the loss or replacement of a laptop from also wiping out configurations, instructions, accumulated context or content generated during work.
AI agent protection in 20 seconds
- Commvault will be able to locate certain AI tools installed on managed corporate devices and protect their local data.
- Recovery covers agent configurations, AI-generated content, prompt libraries, custom instructions and tool-specific context.
- The desktop ChatGPT app can already be detected; protection and recovery for Claude Cowork, Claude Code and OpenClaw will arrive in the coming months.
- The feature is built into Commvault’s existing endpoint protection offerings, with no additional license required according to the company.
- The goal is also to give visibility into possible AI usage outside the processes approved by IT and security teams.
The initiative addresses a problem that can go unnoticed as an organization folds AI tools into daily work. Traditional documents are usually stored on servers, cloud services or corporate systems with backup and recovery mechanisms. Part of the work done with AI agents, however, can end up tied to the employee’s own computer.
That’s where specific configurations, custom instructions, prompt libraries, generated content and other data that help an agent continue a task can end up living. If the machine is lost, damaged or has to be replaced, recovering only the conventional files doesn’t necessarily let you rebuild that working environment.
Commvault also wants to bring that information into the realm of endpoint protection and recovery policies. The company has added this capability to AgentRecover, its offering for recovering enterprise AI environments, including the data and configurations of the tools and connected systems they use.
From file backup to recovering agent context
The difference lies in what an organization considers recoverable. On a developer’s computer, for instance, the value of an AI tool can sit as much in the project files as in the configuration that lets the agent work in a particular way.
Commvault says its IT teams can identify certain AI tool installations on managed devices and protect the configurations stored locally. If something happens to the device, that information can be recovered.
The company specifically mentions AI-generated content, prompt libraries, custom instructions and tool-specific context. The exact scope depends on the local data each application supports, so the announced capability doesn’t amount to a full copy of everything done inside an AI service.
The first tool listed as available for discovery is the ChatGPT desktop app. For Claude Cowork, Claude Code and OpenClaw, Commvault expects general availability of discovery, protection and recovery features in the coming months.
The distinction between discovery and protection matters. Detecting that an application is installed lets administrators know it’s present on corporate devices, but it doesn’t mean all of its data is already protected. For the three tools announced for the coming months, Commvault doesn’t yet present those features as generally available.
AI’s rise is also widening the shadow AI problem
The company ties this capability to another challenge facing IT and security departments: knowing which AI tools employees actually use.
Commvault cites an IBM study in which 80% of the U.S. office workers surveyed said they use AI in their professional work, while only 22% used exclusively the tools provided by their employers. That data comes from an IBM survey published in November 2025 and is used in the announcement to illustrate the gap between approved tools and those that can end up being used independently.
That scenario is usually called shadow AI, referring to the use of artificial intelligence applications outside the processes an organization has established. From a security standpoint, the problem isn’t just knowing which service a person uses. What’s stored locally, which configurations have been created and what data could be left unprotected also matters.
Commvault’s new capability gives security teams a centralized view of certain tools installed on managed devices. If an AI agent application shows up outside approved procedures, the organization can locate and investigate it. That puts Commvault’s move alongside other recent enterprise AI-security launches, such as RSA’s Agent ID platform for controlling what AI agents can do in regulated enterprises.
The company doesn’t claim its system eliminates shadow AI or that it can detect every artificial intelligence tool in existence. The announcement is limited to supported applications and to the managed devices covered by its endpoint protection offering.
Recovery will be one more piece of enterprise AI management
Commvault is building these features into its existing endpoint protection capabilities and says customers don’t need a separate license or an additional SKU to use them. The company also plans to expand the number of supported AI tools in future releases.
The strategy reflects a shift in the kind of information companies have to consider as they bring in AI agents. An agent doesn’t just work with an isolated file: it can depend on instructions, configurations, context and tools that are part of a workflow. If any of those elements is lost, manually rebuilding the environment can be harder than recovering a conventional set of documents.
For now, the offering is at different stages depending on the application. Discovery of the ChatGPT desktop app is available, while discovery, protection and recovery of the local data supported by Claude Cowork, Claude Code and OpenClaw will become generally available in the coming months.
Commvault will also present these capabilities at SHIFT 2026, its event for customers and partners, which the company will hold in Nashville. The announcement places recovery, visibility and cyber resilience as elements of its strategy for managing enterprise environments where the use of AI agents is growing.
For IT departments, the question this raises is concrete: if an agent becomes a regular part of an employee’s work, its configurations and context also carry operational value. Protecting those elements can matter as much as preserving the files the agent works with, though how far that protection actually reaches will depend on each application and the data Commvault can identify and recover.
Frequently asked questions
Which AI tools can Commvault protect?
Commvault is announcing support for local data and configurations tied to Claude Cowork, Claude Code and OpenClaw. Detection of the ChatGPT desktop app is already available, while protection and recovery for the other tools will arrive in the coming months.
What kind of information can be recovered?
The company mentions agent configurations, AI-generated content, prompt libraries, custom instructions and tool-specific context for the supported tools.
Does the feature require buying an additional license?
According to Commvault, these capabilities will be available through its existing endpoint protection offerings, with no separate SKU or additional purchase required.
What’s the point of detecting installed AI tools?
It lets IT and security teams know which supported tools are present on managed corporate devices. Commvault also presents this visibility as a way to help investigate possible AI use outside approved processes.
