AI agents have moved from pilots into production, and they now touch business data directly. That shift creates a problem most enterprises have not solved yet: an agent can change data, applications, and identities at machine speed, and a single bad decision can propagate before anyone notices. Commvault and Zscaler announced a strategic partnership on November 10, 2026, to close that gap — pairing Zscaler’s AI security controls with Commvault’s cyber resilience and recovery so enterprises can govern agents and roll back the damage when something goes wrong.
Why AI Agents Need Governance and a Rollback Plan
The partnership starts from four practical questions security and IT leaders are now asking: which AI tools are employees and developers actually using, what data is being shared with them, which activities should be allowed or blocked, and — if unwanted changes occur — can the business recover quickly? Those questions matter because agents do not behave deterministically. As Commvault’s Chief Technology and AI Officer Pranay Ahlawat put it, a single agent decision can trigger changes across data, applications, and infrastructure on multiple platforms.
That is the core difference between securing a chatbot and securing an agent. A chatbot answers questions; an agent acts. It can delete records, reconfigure systems, move files, and change permissions — and it can do all of that in seconds. Prevention alone is not enough, because some harmful action will eventually slip past the guardrails. Enterprises need both control over what agents may do and a way to undo what they did. Our practical guide to AI agents covers the broader questions teams should ask before putting agents into production, and this partnership is aimed squarely at the governance half of that checklist.
What Each Company Brings to the Partnership
The division of labor is clean. Zscaler Security for AI identifies how AI applications are being used across the enterprise and enforces policy on what users and AI agents can do. Its platform covers discovery of AI assets, secure access to AI tools, and runtime protection for AI applications — the preventive side of the equation.
Commvault brings the recovery side. Its AI and cyber resilience capabilities help enterprises recover data, applications, and identities after attacks, disruptions, or outages. The idea is a continuous loop: Zscaler sees and controls the agent activity, Commvault restores whatever the activity damaged. As Zscaler’s Dhawal Sharma, EVP of AI Security and Strategic Initiatives, said in the announcement, Zscaler secures AI use across the enterprise while Commvault strengthens recovery and resilience — and together they give organizations the controls to manage risk and protect sensitive data.
AI Discovery: Finding the Shadow AI First
The first planned use case is comprehensive visibility and shadow AI governance. Before an enterprise can govern its agents, it has to know they exist — and many of them are invisible to IT. Employees spin up AI tools, developers wire agents into workflows, and teams adopt AI-powered SaaS apps without a ticket ever being filed.
The partnership plans to advance enterprise-wide discovery that keeps continuous visibility into active AI applications, models, and tools. That matters because shadow AI is where the riskiest agent activity usually starts: an unsanctioned tool with access to a production database, or an agent built on a personal account with no audit trail. Discovery is the unglamorous foundation the rest of the governance story depends on, and it is also where most enterprises are weakest today.
Runtime Governance: Guardrails While Agents Work
The second use case is runtime governance — active guardrails that flag unsanctioned AI activity and prevent sensitive data exposure while agents are running. This is policy enforcement at the point of action, not a quarterly review: deciding in real time which tools an agent may call, which data it may touch, and which actions require a human in the loop.
This is the same direction the rest of the industry is moving. Cisco’s Claude Mythos red teaming for AI Defense, announced the same week, tests agents with frontier reasoning before they ship; the Commvault–Zscaler approach governs them while they run and cleans up afterward. Together, the two announcements sketch an emerging standard for the agentic enterprise: test before deployment, govern during operation, and recover after failure. The governance angle also echoes the sovereignty concerns behind Cohere’s sovereign AI partnership in Singapore, where control over where AI runs and what it touches is the whole point.
AgentRecover: Rolling Back What Agents Break
The third use case — resilient workload security and automated recovery — is where Commvault’s new AgentRecover capabilities come in. Announced alongside the partnership, AgentRecover is designed to audit agent activity, restore agents, and recover the data and systems they affected across environments, with support for Amazon Bedrock AgentCore and Microsoft Foundry, the platforms enterprises use to build and operate production agents.
AgentRecover works in three steps. First, it discovers agents and maps the data, applications, identities, configurations, and infrastructure they interact with, exposing protection gaps. Second, its auditing functions establish what an agent did and what it may have affected, so teams can prioritize recovery work. Third, recovery covers both the agent and the affected resources, concentrating on exactly what the agent changed — and it can be combined with Commvault’s identity recovery and change auditing to trace what agents did, what data they changed, and which identities they acted through. Early access for Microsoft Foundry is targeted for November 2026, with Amazon Bedrock AgentCore following in December 2026.
What to Watch Next
The partnership is a plan, not a finished product: the companies described the use cases as planned collaboration areas rather than a generally available joint offering. The real test is execution — whether discovery, governance, and agent recovery become working capabilities for joint customers, and how quickly. The companies will showcase the work at SHIFT Virtual on November 18, 2026.
For enterprises, the practical takeaway is to treat agent governance as a resilience problem, not just a security policy. Ask which agents are running in your environment, what they can change, and whether you can roll back their actions — and build the answer around the same governance thinking behind Snowflake’s agentic enterprise stack, where governed agents with shared business context are becoming the baseline. The vendors are converging on the same conclusion: the agent era needs a recovery plan, not just a permission slip.
