Commvault and Zscaler pair AI controls with a plan for recovering from agent mistakes

guthlabs.newsroom | AI Security | November 10, 2026
Commvault and Zscaler announced a strategic partnership on November 10 aimed at combining controls over enterprise AI use with recovery after harmful agent activity. The companies plan to collaborate on discovering AI systems, enforcing policy during operation and recovering from unwanted changes. Commvault partnership announcement
The proposed division of labor joins Zscaler’s visibility and policy enforcement with Commvault’s ability to recover data, applications and identities. Listed opportunities include identifying unsanctioned AI use, protecting sensitive information and supporting clean recovery following risky agent actions. The announcement describes areas of collaboration, rather than a named joint product with a general-availability date. Commvault partnership announcement
That distinction matters for buyers. A partnership can establish a useful technical direction without yet answering which environments are supported, how the products exchange incident information or who takes responsibility when an automated response fails. Those details determine whether two complementary capabilities become one dependable operating process.
The companies point customers to Commvault’s SHIFT Virtual event on November 18 for more information. The release does not disclose financial terms or quantified results from a joint deployment. Commvault partnership announcement
Recovery has a different job from prevention
An agent with permission to change systems creates two separate security problems. An organization must decide whether an action should be allowed, then remain able to investigate and repair the consequences if something goes wrong. Blocking a suspicious request and restoring a damaged database are related, but they are not interchangeable tasks.
The business case for connecting them is strongest when the security event can be tied to a specific operational change. An alert becomes more useful if responders can see which identity acted, which resources changed and which recovery point remains trustworthy. Without that connection, teams may still have to reconstruct the sequence manually.
Commvault separately announced new AgentRecover capabilities on the same day, covering audits of agent activity and recovery of agents and affected resources. It targeted Microsoft Foundry early access for November and Amazon Bedrock AgentCore for December. Those targets belong to the AgentRecover announcement, not to a released joint offering with Zscaler. Commvault AgentRecover announcement
This separate timetable illustrates why enterprise buyers should distinguish a product roadmap from a partnership roadmap. The availability of one component does not establish that every promised integration is ready, nor that every application an agent touches is recoverable through it.
Test the limits of rollback
A meaningful evaluation should start with concrete failure scenarios. Can the system identify a mistaken configuration change, preserve the evidence and restore the affected service without undoing legitimate work performed afterward? Can responders stop an agent before restoration starts, so the same process does not immediately recreate the damage?
Some consequences also extend beyond a recoverable system. Restoring a database does not necessarily retract an externally delivered message or reverse another organization’s response. That is a reason to retain preventive controls and explicit authority limits, even where strong backups exist.
Commvault and Zscaler are proposing a connection between those two responsibilities. The next evidence buyers need is the scope of the integration, its release status and results from realistic recovery exercises. Until then, the announcement is a direction for joint development, not proof that agent-driven incidents can be universally undone.
Sources and citations
Revision history
Published .