Role-based access control determines whether AI agents are trustworthy

Companies scrambling to bring AI agents into production are finding that role-based access control, not model selection, determines whether these agents can be trusted. Rising inference costs, privacy concerns and sovereignty requirements are also drawing more AI workloads back into enterprise data centers.

This shift favors companies that have spent years managing sensitive data at scale. Cloudera Inc. relies on Cloudera Anywhere Cloud, a hybrid data and AI platform that spans public clouds and on-premises environments. According to Leo Brunnick (pictured), Cloudera’s chief product officer, the company has long been strong in governance, security and scalability, but weaker in usability – a gap the new platform aims to close.

“It turns out that the model ultimately isn’t going to be what separates your success from the success of your competitors,” Brunnick said. “The things that will separate you are, first and foremost, your unique data. Your unique data and your unique processes, your alpha – what makes you a company.”

Brunnick spoke with theCUBE’s Rebecca Knight and Paul Gillin at EVOLVE26 during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed why AI agents must follow users’ data permissions, the forces pushing AI workloads back into enterprise data centers, and how packaged plans and partnerships can simplify private AI adoption. (*Disclosure below.)

Why role-based access control now extends to AI agents

Customers rarely want a new platform for its own sake, so Cloudera instead sells solutions to specific problems. Its packages and expanded partnerships with Mistral AI SAS, Nvidia Corp. and Vast Data Inc. target private AI use cases such as fraud detection. Mistral, which specializes in open-weight models for on-site inference, also gives Cloudera something it has been missing in the AI ​​market.

“Cloudera already had the capabilities. What we lack is great credibility around AI. People don’t think of us as an AI company,” Brunnick said. “With Mistral we can easily increase our capabilities and increase our credibility astronomically.”

The more difficult work lies in governance. To be useful, agents need to be able to see data across the organization, so everyone needs to know who the user is and what that user is allowed to see, Brunnick explained.

“Because the agent needs to see everything in the company, they need to know who you are and what permissions you have,” Brunnick said. “That’s what they need to get right so they can actually put things into production that won’t blow up in their face.”

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