Cohesity has received a US patent for technology used in its generative AI platform, Gaia. It says it is the first data protection vendor to secure a patent for this approach.
The patent covers a method for using retrieval-augmented generation with data held in backup systems. This allows organisations to search and use secondary data for AI applications without moving it into separate environments, while preserving existing governance, compliance, security and access controls.
Secondary data typically includes copies of files, emails, databases and virtual machines kept for recovery and retention. Although companies often hold years of this information, much of it has not been used for AI projects because moving sensitive records into new tools can create extra risk, duplicate data sets and weaken control over access.
Cohesity said its patented method enables large language models to work against protected secondary data through a semantic search layer. In practice, that means organisations can use backup and recovery data as a knowledge source for AI systems while keeping the original information in place.
The patent is titled Data Retrieval Using Embeddings for Data in Backup Systems. Cohesity named Gregory Statton, Sanjay Poonen, Mohit Aron and Apurv Gupta as inventors.
The move comes as businesses look for ways to use generative AI on internal data without creating new stores of sensitive information. For companies in regulated sectors, this has become central to AI adoption, as data residency, internal controls and audit requirements can limit how information is shared across systems.
Data in place
Cohesity argues that the key distinction in its design is that AI queries run against data already held in secondary storage, rather than requiring information to be copied into a separate AI repository. It says this can help reduce data sprawl and limit the exposure of sensitive material.
That position reflects a wider debate in enterprise technology over whether AI systems should be built around newly assembled data lakes or existing operational and archival stores. Vendors have increasingly tried to reassure customers that AI tools can be introduced without undermining established security policies.
In this case, Cohesity is tying that argument to intellectual property as well as product design. It says the patent validates the architectural approach behind Gaia, its enterprise generative AI product.
"Protected recovery data is a goldmine. It is an organisation's most important, complete and trusted repository of enterprise information and institutional knowledge. Yet it remains among the most underutilised," said Sanjay Poonen, chief executive officer and president of Cohesity.
"This patent reflects years of foundational engineering work to change that with a security-first architecture for enterprise AI. Cohesity Gaia applies AI directly to that data, without forcing organisations to move or duplicate sensitive information, unlocking insights while maintaining the governance, access controls and cyber resilience they depend on. No other platform delivers these advantages today," Poonen said.
Commercial focus
Gaia is available as part of Cohesity's Data Cloud platform. The company positions the product as a way for enterprise teams to search historical data and build AI-driven workflows using information already stored for backup and recovery.
The patent could strengthen Cohesity's position in a market where suppliers of data protection, storage and security software are all trying to define a role in enterprise AI. Backup platforms have traditionally been sold on resilience and recovery, but vendors increasingly argue that the same repositories also contain valuable institutional knowledge.
For customers, the practical question is whether those archives can be used without creating additional operational or compliance burdens. That is especially relevant for public sector bodies and heavily regulated industries, where restrictions on data handling can slow or block AI roll-outs.
Gregory Statton, one of the inventors, said the work was intended to change how companies think about existing backup data.
"The patent behind Gaia represents a fundamental shift in how enterprises think about the data they already hold," said Greg Statton, CTO and VP, APJ, Cohesity. "For years, organisations have sat on vast reserves of rich, governed but largely untapped secondary data. My thinking, when working with the team on its creation, was to change all that. By enabling AI-driven insights directly from backup data, without moving or duplicating it, we've built something genuinely unique: a patented RAG architecture that keeps data in place, maintaining all of the security and process management advantages that come with that, while unlocking its full value."
He added that customer interest has shifted from experimentation to practical deployment.
"We knew at the time this capability would only increase in value for our customers, and today, as agentic AI moves from experimentation into core business workflows faster than we could have predicted, that foundation is an even more important differentiator," Statton said. "Customers today aren't just asking how to use AI, they're asking how to use it safely, compliantly and at scale. Gaia was built to answer that question, and we're seeing real, measurable value being delivered to customers now running agentic workflows on data they previously couldn't use. More importantly, that shift in customer need is only going to accelerate."
A customer from the Dutch public sector also linked the approach to data control requirements.
"In evaluating enterprise AI approaches, preserving our security posture and ensuring sovereign, on-premise control as a Dutch government institution were critical objectives," said Patrick Ringelberg, domain data centre architect and AI, Ministry of Infrastructure and Water Management (Rijkswaterstaat).
"Cohesity's approach was the only one that made AI viable using our existing backup data as the foundation. The fact that this approach is now patented reinforces how differentiated it is," Ringelberg said.