Commvault presents at AI Infrastructure Field Day 5

Building AI Resilience from a CISO’s Lens with Commvault

Watch on YouTube
Watch on Vimeo

AI infrastructure introduces a new class of risk: interdependent data, models, identities, and pipelines can fail out of sync, leaving systems operational but no longer trustworthy. At the same time, a growing CIO/CISO disconnect is creating unclear ownership and slowing response when trust is lost. Commvault aims to close this gap with a shared control plane approach that governs access, protects the full AI stack, and recovers systems to a clean, coherent, and trusted state. Chris Bevil, a former CISO, set the stage for the presentation by emphasizing the “people” aspect alongside processes and technology, drawing insights from his interactions with CISOs globally. He highlighted the critical need to address the AI-specific resilience gap, understand how AI operates, establish robust governance, and ensure a trusted path to recovery.

A core focus of the presentation from the CISO’s lens is the “ownership gap” in AI resilience. It’s no longer solely the responsibility of IT or operations; instead, it’s a collaborative effort involving security, finance, and specialized AI teams. The paramount concern is “trust”: can the recovered systems and data be trusted to be clean and coherent? Bevil articulated the CISO’s evolving role, shifting from solely security to encompassing data recovery and business continuity, stressing the importance of enabling AI innovation rather than being a “department of no.”

To address these challenges, Commvault’s strategy emphasizes unifying efforts across diverse business units. Without a shared understanding and strategy, teams risk becoming siloed, hindering rapid and effective decision-making during an incident. The company’s approach to building AI resilience involves securing the entire AI infrastructure, including the data foundation, pipeline and processing, AI data stores, and the modern lifecycle, all orchestrated through a comprehensive control plane. This integrated approach ensures that when recovery is needed, the entire AI stack can be restored to a proven, trustworthy state, thereby maintaining operational integrity and confidence.

Personnel: Chris Bevil

Re-engineering People, Process, & Tools for AI Resilience with Commvault

Watch on YouTube
Watch on Vimeo

AI infrastructure introduces a new class of risk: interdependent data, models, identities, and pipelines can fail out of sync, leaving systems operational but no longer trustworthy. At the same time, a growing CIO/CISO disconnect is creating unclear ownership and slowing response when trust is lost. In this session, Commvault will show how to close that gap with a shared control plane approach: governing access, protecting the full AI stack, and recovering systems to a clean, coherent, and trusted state.

The presentation elaborated on how AI has significantly complicated traditional IT stacks, transforming previously linear and human-paced data flows into chaotic, nonlinear, multipath, stateful, and agentic processes. This expansion has led to an explosion of data copies, identities, and threat vectors, creating immense confusion regarding ownership and responsibility for threat response, patching, and control implementation. Commvault has responded by shifting its focus from mere data protection to a comprehensive “resilience operations” (ResOps) methodology over the last five years. This strategic pivot aims to help organizations recover not just quickly but also cleanly and reliably from adverse events, addressing a critical board-level concern that extends beyond traditional disaster recovery and cybersecurity to encompass new risks such as rogue AI agents and sensitive data handling within the AI ecosystem.

The speaker further detailed the new, complex recovery dependencies introduced by AI, including the critical importance of source, sensitivity, access, table, embedding, vector index, and feature versions. These elements must be recovered in sync for AI applications to function correctly and maintain trust. Simply restoring individual workloads, as was common in the past, is no longer sufficient; instead, the entire AI system must be brought back to a consistent, coherent, and trusted state. Commvault addresses this by developing reusable, agnostic frameworks designed to adapt to the rapidly evolving AI landscape, including new data formats and vector databases. This proactive approach provides a crucial “safety net,” automating recovery, ensuring data quality, and offering auditability and explainability, especially important in mitigating risks from shadow AI and the rapid, often unpredictable actions of AI agents where human-speed responses are inadequate.

Personnel: Chris Bevil, Michael Fasulo

Governing Data for AI Through a New Control Plane with Commvault

Watch on YouTube
Watch on Vimeo

AI infrastructure introduces a new class of risk: interdependent data, models, identities, and pipelines can fail out of sync, leaving systems operational but no longer trustworthy. At the same time, a growing CIO/CISO disconnect is creating unclear ownership and slowing response when trust is lost. In this session, Commvault will show how to close that gap with a shared control plane approach: governing access, protecting the full AI stack, and recovering systems to a clean, coherent, and trusted state. Jose Gomez, Global Field CTO at Satori (acquired by Commvault in mid-2025), highlighted the company’s evolution beyond traditional backups, with a focus on AI data access governance. This includes managing structured data across environments such as SQL Server, Snowflake, and Databricks, and overseeing its use to train AI models and LLMs. The core objective is to govern data *before* AI agents or humans access it, establishing a new governance control plane to mitigate risks in a rapidly evolving AI landscape.

The new approach to governance centers on a proxy-based application that acts as a Data Access Controller (DAC) and an LLM gateway. This system intercepts data requests from both human users and AI agents, applying real-time policies. Leveraging APIs and a Kubernetes platform, it integrates with identity providers such as Active Directory to manage user access and enforce granular policies, including data classification and redaction of sensitive information. This ensures that even when AI agents act on behalf of users, data access is controlled, preventing sensitive information from entering RAG databases or being inadvertently captured in training logs. The platform offers flexible deployment options, including a fully managed cloud service or customer-hosted Kubernetes, addressing data sovereignty and regional compliance requirements like GDPR by keeping data processing within specified geographic boundaries.

Beyond governance, Commvault emphasizes “trusted recovery” for the entire AI stack. Recognizing that AI applications involve complex interdependencies among data, models, identities, pipelines, and underlying infrastructure, traditional data recovery methods are insufficient. The platform aims to provide comprehensive data protection across all these layers to ensure that, in the event of a compromise, the entire AI system can be quickly restored to a coherent, trusted, and functional state. Commvault positions itself as a critical layer in cyber resiliency, offering visibility and guardrails to manage the dynamic AI environment, balancing business efficiency with robust security and compliance amid rapidly evolving AI models and applications.

Personnel: José Gomez

The Path to Trusted AI Application Recovery with Commvault

Watch on YouTube
Watch on Vimeo

AI infrastructure introduces a new class of risk: interdependent data, models, identities, and pipelines can fail out of sync, leaving systems operational but no longer trustworthy. At the same time, a growing CIO/CISO disconnect is creating unclear ownership and slowing response when trust is lost. In this session, Commvault will show how to close that gap with a shared control plane approach: governing access, protecting the full AI stack, and recovering systems to a clean, coherent, and trusted state.

Commvault addresses these challenges by enabling the discovery and protection of entire AI applications as holistic entities, rather than individual workloads. Their Azure Discovery module, for example, identifies all resources that comprise an AI application (e.g., front end, vector database, AKS, data lake) and enables policy-based protection using rules or tags, ensuring consistent and scalable management. A critical component is the “Clean Room” feature, which creates isolated, ephemeral environments for continuous testing of recovery and patching processes. This allows organizations to build muscle memory for trusted recovery and test vulnerability fixes or compensating controls without impacting production. The core “run book” orchestrates these processes by defining foundational and custom steps, enabling recoveries to specific points in time, integrating threat scans, and offering full customization for scenarios such as patching or disaster recovery.

The presentation also highlights Commvault’s “ResOps” (Resilience Operations) framework, emphasizing that achieving resilience in AI infrastructure requires a concerted effort across people, processes, and tools. Commvault integrates its own AI assistant, Arlie, throughout its interface to enhance operational excellence. Arlie synthesizes insights, provides root cause analysis, offers advisory recommendations (e.g., for protection gaps), and guides users through complex recovery decisions to ensure the cleanest possible restoration, leveraging metadata and threat signals. Furthermore, Commvault’s platform (MCP) can interface with external large language model harnesses such as Claude or Copilot, enabling users to manage infrastructure deployment and Commvault protection policies seamlessly through these tools, effectively creating a “headless” operational experience. This comprehensive approach ensures that AI data, models, identities, and infrastructure are protected, access is governed, and recovery is consistently clean and trusted.

Personnel: Michael Fasulo