Protect and secure your data from cyber attacks
Data Protection
Data Security
Data Insights
The 5 Steps to Cyber Resilience
Cloud & SaaS
Enterprise
Industries
Most major data protection vendors have started integrating artificial intelligence (AI) capabilities in their platforms. Backup repositories tend to be among the most complete, well-governed data sets an organization owns, and AI systems function more efficiently when they have access to trustworthy data. The difference between vendors is in the manner they transfer this data in AI processes while preserving any controls in place.
This article compares three platforms taking on that problem: Cohesity Gaia, Rubrik Annapurna, and Commvault Data Activate. Each takes a distinct architectural approach to making protected enterprise data usable for AI without creating new security risks.
AI data activation makes backup and protected data usable by AI models, analytics platforms, and data science workflows, without exposing sensitive data to systems that shouldn't have it.
“AI gets more useful with the more data you put into it, but for someone in a regulated industry like banking or public sector, you can't just bring that data to external AI tools,” says Brian Chang, AI and Data Product Strategist at Cohesity. “It's sensitive, it has policies attached to it, and moving it breaks those policies. What we provide is a governed data layer on top of your protected data so you can activate it safely in your AI tools without breaking those controls.”
At a high level, the work involves scanning protected data to identify what's available and what's sensitive, preserving access controls and compliance policies, and getting curated datasets into the formats that AI teams need. Each vendor approaches that pipeline in their own way, and the architectural choices carry real implications for deployment, cost, and how well-governed the data stays end to end.
Cohesity Gaia is an AI-powered tool that converts backup data from a static recovery repository into a searchable asset. Instead of leaving backup copies dormant, Gaia uses retrieval-augmented generation (RAG), letting you use your historical, protected data to power applications and fuel production AI initiatives.
RAG lets users query backup data using natural language, with the system answering by retrieving relevant snapshots and metadata. For example, you could ask a question like, “Show me all customer support tickets referencing outage X” directly against protected backup data sets. This helps teams reuse compliant, secure data without impacting operations by querying production data.
Gaia is designed to run in strictly controlled environments, including on-premises deployments and air-gapped architectures where backup data is isolated from the public internet. This matters for organizations with stringent data sovereignty needs or regulated workloads. Bringing AI to the data inside secure zones (instead of moving data out to the AI and risking exposure) enables Gaia to help enterprises keep ownership of their information while still unlocking AI-driven insights.
Gaia integrates with broader enterprise AI workflows through the Model Context Protocol (MCP). MCP gives external AI tools a standardized, read-only way to search and retrieve governed context from backup data — no write access, no raw-data exposure. Backup datasets become governed context sources for internal AI assistants, development tools, and analytics agents without exposing the raw data, letting organizations connect Cohesity backup data to their existing AI stack while maintaining data security and access controls.
Rubrik Annapurna positions itself as an unstructured data layer for enterprise data intelligence platforms. Instead of moving protected data into a separate environment for AI use, Annapurna scans and catalogs unstructured data in place across distributed systems and publishes a queryable index into a lakehouse. AI and data science teams can then pull the subsets they need.
Annapurna's core differentiator is that it avoids duplicating entire data estates. Traditional approaches require organizations to extract, transform, and load full environments into a data lake before AI teams can get access. Instead, Annapurna builds a metadata catalog across distributed file systems and lets data engineers query the index to find specific targets. Only the selected subset gets staged for downstream workflows, and pipeline costs scale with what AI actually consumes rather than the size of the full estate.
Annapurna builds on Rubrik Security Cloud and its Zero Trust foundation to attach verifiable lineage and versioning to staged files, tracking provenance from source through AI output. Native source-system access controls are preserved directly within the catalog, allowing downstream platforms to enforce permissions without adding an extra governance layer. Rubrik positions this as support for GDPR and other compliance programs that require demonstrable data lineage. On the integration side, Annapurna deploys alongside existing storage and lakehouse environments without new agents or dedicated hardware, and publishes its queryable catalog directly into Databricks Unity Catalo
Commvault Data Activate takes a curation-first approach to AI data activation. Built within Commvault Cloud, Data Activate gives users a governed workspace for discovering relevant datasets within their protected backups, applying classification and redaction policies, then exporting curated data in AI-friendly formats. The idea is to have trusted datasets with compliance guardrails in place before any data reaches an AI pipeline.
Data Activate includes built-in classification and sensitivity tagging that lets teams identify high-value or regulated data within backups. Before any data leaves Commvault Cloud, teams can apply redaction policies to exclude PII or other sensitive information from datasets. Every action is tracked through audit trails, and the system operates within Commvault Cloud's zero-trust architecture.
Rather than one-time bulk exports, Data Activate continuously publishes updated and vetted datasets to keep AI pipelines in sync with trusted data. Exports use open standards like Apache Parquet and Apache Iceberg, making datasets consumable by platforms like Microsoft Azure and Snowflake without custom ETL pipelines. Commvault connects Data Activate to the broader AI ecosystem through its MCP server, letting AI assistants and automation tools query data and initiate recovery through conversational interfaces.
Here’s a high-level comparison of the three platforms across six key dimensions:
Dimension
Cohesity Gaia
Rubrik Annapurna
Commvault Data Activate
Primary approach
RAG-based conversational search on backup data
Scan-in-place cataloging with demand-driven pipelines
Governed dataset curation with continuous publishing
Core differentiator
Natural language queries against historical backup datasets
Metadata index avoids full-estate duplication, and costs scale with consumption
Built-in classification, redaction, and policy enforcement before data leaves the platform
Deployment emphasis
On-premises, air-gapped, and hybrid environments
Alongside existing lakehouse and storage infrastructure
Within Commvault Cloud (SaaS)
Data delivery model
Conversational retrieval via RAG
Staged subsets published to lakehouse
Curated datasets exported in Parquet and Iceberg formats
AI ecosystem integration
MCP for secure context sharing with enterprise AI tools
Google Agentspace and native lakehouse connectors (forthcoming)
MCP server and integrations with Azure, Snowflake, ChatGPT Enterprise, Claude
Governance model
Inherits existing backup permissions and access controls carry forward
Zero Trust chain of custody with source-system ACLs preserved in catalog
Classification, sensitivity tagging, redaction, and audit trails within Commvault Cloud
Gaia, Annapurna, and Data Active all turn protected data into a governed resource for AI, but each makes distinct architectural tradeoffs. The best fit depends on where your data lives and how your AI teams want to consume it.
Most enterprise use cases with AI doesn't need the best model. As organizations operationalize AI, the differentiator is going to be their data, not which model they deploy."
rather than building export pipelines. Its RAG capabilities let teams ask natural language questions against historical datasets, making backup data accessible to analysts and AI workflows without a separate platform. On-premises and air-gapped deployment keeps sensitive data under strict control, making Gaia the strongest fit for regulated industries and sovereign AI use cases where data cannot leave a controlled environment. Combined with Cohesity's broader data security and resilience platform, Gaia positions backup data as both a recovery resource and a governed knowledge base for enterprise AI.
Rubrik Annapurna leads where organizations need to prepare large unstructured file estates for data science and AI model training at scale. The scan-in-place architecture avoids the cost and complexity of duplicating entire environments into a data lake, and the demand-driven pipeline model means infrastructure costs track actual AI consumption. For companies with lakehouse architectures or Google Cloud's AI platform, Annapurna helps get data ready for AI.
Commvault Data Activate leads for organizations that need tight control over what leaves the backup environment. The classification and redaction workflow guarantees that datasets are scrubbed and approved before they reach any AI platform, which is why Data Activate is a good fit for teams that need to perform a compliance check before transferring any data.
The value of any data activation capability depends on how complete and well-governed your protected data is and how accessible it is to the tools your teams already use. Gaps in backup coverage or inconsistent access controls limit what any AI layer can deliver.
Cohesity starts with a unified data resilience and data backup and recovery platform, then layers Gaia on top to activate that protected data for AI. By combining backup, security, and data services on a single platform, Cohesity positions backup data as both a recovery safety net and a governed, queryable foundation for enterprise AI.
