Project-by-project security reviews can’t keep pace with enterprise AI agents
For architects, the unresolved question is interoperability. Tokenized data may remain analytically useful, but only if embeddings, joins, search quality and agent decision logic still work with substituted values. In practice, teams will need to test which workflows tolerate tokenization, which require format-preserving approaches, and where selective reveal is unavoidable. Otherwise, enterprises risk building agent pipelines that are compliant on paper but too degraded to deliver business value.
The source also points to an operating-model issue: project-by-project review does not scale when agent use cases proliferate. Pattern approval only works if those patterns are concrete and enforceable. Useful examples include preapproved retrieval architectures, standard controls for vector databases, logging rules that prevent prompt leakage, and reference integrations between data platforms, policy engines and agent frameworks.
For IT leaders, the practical takeaway is to treat AI data protection as platform engineering. Standardize tokenization, access mediation and auditability as reusable services early, or every new agent initiative will recreate the same governance bottleneck with higher operational risk.
Enterprises racing to put AI agents into production are finding that tokenization and other data protection methods now decide how fast those projects move.
But sensitive data fed into models is hard to pull back — and security reviews built for a slower era are stalling deployments. The challenge is shifting from locking data down to keeping it usable once it is protected. That tension is playing out across Cloudera Inc.’s partner ecosystem, where security is becoming a prerequisite for AI value rather than a final gate, according to Diego Loureda (pictured), vice president of customer success and presales at Protegrity USA Inc.
“Data leaks are still a thing. If you put sensitive data on a model, if you put sensitive data on a vector store or in the memory of an agent that is going to propagate to other agents, good luck removing that sensitive data,” Loureda said. “There’s no way you’re going to be able to do that.”
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 along with Protegrity being named Cloudera’s Emerging Technology Partner of the Year. (* Disclosure below.)
Tokenization keeps protected data usable for AI
Traditional encryption can lock data so tightly that analysts and models can no longer use it. Protegrity centers its approach on tokenization, which swaps sensitive values for substitutes instead of relying on a mathematical algorithm, Loureda explained. That design also matters as quantum computing threatens today’s cryptography.
“Our main method of protection, that is tokenization, it’s already quantum ready,” Loureda said. “The reason behind that is because there is no mathematical algorithm behind it. It’s all based on replacement in memory.”
Protection alone doesn’t solve the speed problem, however. In Original Postrotegrity-research-reveals-the-hidden-ai-friction-tax-holding-back-enterprise-ai" shape="rect">research Protegrity commissioned from Enterprise Management Associates Inc., 83% of leaders reported project delays, Loureda noted.
“The main problem with governance is that companies are approving projects and that will not scale in the age of AI,” Loureda said. “They have to approve patterns.”
The same study found 82% of leaders push agents into production in a diminished state, stripped of the tools and data that would make them useful, he added. Controls such as tokenization let security teams clear the way for those agents instead, according to Loureda.
“Security has always been the business of no, the department of no. That changes now,” he said. “These days security is the team that can enable AI to give you value.”
Here’s the complete video interview, part of SiliconANGLE’s 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 content on theCUBE or SiliconANGLE.)
Project-by-project security reviews can’t keep pace with enterprise AI agents
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.