Security gatekeeping slows enterprise AI agent rollouts as data protection becomes the bottleneck

Organizations moving AI agents into production are discovering that how quickly they can ship depends heavily on tokenization and similar data safeguarding techniques. Yet the data fed into these models creates a lasting problem—once sensitive information enters a model, vector store, or agent memory, extracting it becomes nearly impossible. Meanwhile, security approval processes built for an earlier, slower era of software development are creating bottlenecks that delay these initiatives. The real shift underway is moving security from a purely defensive posture to one that preserves data utility after protection is applied. This dynamic is becoming visible across Cloudera Inc.'s partner network, where security now functions as a prerequisite for AI value creation rather than a final checkpoint, according to Diego Loureda, 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. There's no way you're going to be able to do that.

Diego Loureda, Protegrity USA Inc.

Loureda shared these insights during an exclusive conversation with theCUBE's Rebecca Knight and Paul Gillin at Cloudera EVOLVE26. The discussion also covered Protegrity's recognition as Cloudera's Emerging Technology Partner of the Year.

Tokenization enables protected data to remain useful for AI

Standard encryption methods can render data so inaccessible that neither analysts nor AI models can work with it effectively. Protegrity's strategy centers on tokenization, which replaces sensitive information with stand-in values rather than applying mathematical algorithms, Loureda explained. This approach also carries implications for the quantum computing era, as traditional cryptographic methods face emerging threats.

Our main method of protection, that is tokenization, it's already quantum ready. The reason behind that is because there is no mathematical algorithm behind it. It's all based on replacement in memory.

Diego Loureda, Protegrity USA Inc.

Yet protection mechanisms alone cannot resolve deployment velocity challenges. Research commissioned by Protegrity through Enterprise Management Associates Inc. revealed that 83% of organizational leaders reported experiencing 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. They have to approve patterns.

Diego Loureda, Protegrity USA Inc.

The same research indicated that 82% of leaders deploy agents in limited configurations, stripped of capabilities and data access that would maximize their effectiveness, Loureda added. Tools like tokenization can shift this dynamic by enabling security teams to authorize broader agent functionality rather than restricting it.

https://www.youtube.com/embed/-K-nAaar5mk?feature=oembed

Security has always been the business of no, the department of no. That changes now. These days security is the team that can enable AI to give you value.

Diego Loureda, Protegrity USA Inc.