Tokenization helps companies bring AI agents into production

Companies struggling to bring AI agents into production are finding that tokenization and other data protection methods are now determining how quickly these projects move forward.

But sensitive data fed into models is difficult to retract—and security checks designed for a slower era delay deployments. The challenge is to stop locking data and keep it usable after protection. According to Diego Loureda (pictured), vice president of customer success and presales at Protegrity USA Inc., this tension exists across Cloudera Inc.’s partner ecosystem, where security is becoming a prerequisite for AI value rather than the final gate.

“Data leaks are still a thing. If you put sensitive data in a model, if you put sensitive data in vector storage or in an agent’s storage to be passed on to other agents, good luck removing that sensitive data,” Loureda said. “There’s no way you’re going to succeed.”

Loureda spoke with theCUBE’s Rebecca Knight and Paul Gillin at Cloudera EVOLVE26 during an exclusive interview on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed data security and Protegrity being named Cloudera’s Emerging Technology Partner of the Year. (*Disclosure below.)

Through tokenization, protected data remains usable for AI

Traditional encryption can lock data so tightly that analysts and models can no longer use it. Protegrity focuses its approach on tokenization, which involves exchanging sensitive assets for surrogates rather than relying on a mathematical algorithm, Loureda explained. This design is also important because quantum computers threaten today’s cryptography.

“Our main protection method, tokenization, is already quantum ready,” Loureda said. “The reason for this is that there is no mathematical algorithm behind it. Everything is based on replacement in memory.”

However, protection alone does not solve the speed problem. In Protegrity research commissioned by Enterprise Management Associates Inc., 83% of executives reported project delays, Loureda found.

“The main problem with governance is that companies approve projects and these are not scalable in the age of AI,” Loureda said. “You have to approve samples.”

The same study found that 82% of executives are pushing agents into production in a degraded state, without the tools and data that would make them useful, he added. Controls like tokenization instead allow security teams to clear the path for these agents, Loureda said.

“Security has always been No’s business, No’s department. That’s changing now,” he said. “Today, security is the team that can enable AI to bring you value.”

Here’s the full video interview, part of SiliconANGLE and theCUBE’s coverage of Cloudera EVOLVE26:

(*Disclosure: TheCUBE is a paid media partner for the EVOLVE26 2026 event. Neither Cloudera, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over theCUBE or SiliconANGLE content.)

Photo: SiliconANGLE