AI Business

Data governance at machine speed: How HPE and Daxa are securing AI agents in production

As enterprises deploy AI agents with access to sensitive data, traditional security tools are falling short. New approaches like semantic caching are moving governance into the cognitive layer where agents operate.

·3 min read
Agents want the crown jewels: HPE and Daxa push governance down to the data layer
Agents want the crown jewels: HPE and Daxa push governance down to the data layer

When AI agents move from pilot deployments to production systems, they gain the ability to read, modify and remove critical business records. This escalation transforms data access control into the primary security concern for agentic AI, according to Huseni Saboowala, co-founder and chief executive officer at Daxa Inc.

For agents to really get to value, they need to be able to interact with your real enterprise data, and that includes your crown jewels, your intellectual property, your customer data in your [customer relationship management] and your proprietary trading algorithms. That creates an unprecedented level of risk for enterprises because the agents are not just reading data; they're also able to now go manipulate data at scale.

Huseni Saboowala, Daxa Inc.

Saboowala and Robin Braun, vice president of AI business development at Hewlett Packard Enterprise Co., discussed these challenges during SC26 coverage on theCUBE, SiliconANGLE Media's livestreaming studio. Daxa participates in HPE's Unleash AI program, which delivers pre-validated software running on HPE Private Cloud AI.

Governance must operate where agents think

Conventional security infrastructure—firewalls, data loss prevention systems and identity management platforms—was designed for human users and traditional applications. These tools cannot effectively monitor the cognitive layer where AI agents operate, Saboowala explained.

Those agents are operating, though, at layer 8, what we call the cognitive layer. This is where you have the prompts, the responses, the retrievals and the tool calls happening. Traditional tools really are not operating at this layer.

Huseni Saboowala, Daxa Inc.

Daxa's approach centers on reasoning-based access controls paired with agent systems that detect unusual patterns. HPE has taken a different path by embedding governance capabilities directly into its Private Cloud AI infrastructure from the outset, according to Braun.

Being able to build in that governance from day one allows you to understand what is it going to look like so that … you're not bolting it on later. Instead, it's a part of the workflow.

Robin Braun, Hewlett Packard Enterprise Co.

Timing proves critical in agent governance. Offline enforcement mechanisms arrive too slowly for systems operating at machine speed. Instead, controls must execute inline through semantic caching, specialized small language models and a distributed data plane architecture, Saboowala noted. A regional bank that implemented Daxa alongside HPE infrastructure has now granted its agents access to source code repositories and regulated customer information.

https://www.youtube.com/embed/0kfurS_z5jg?feature=oembed

I believe that the winners in this agent era are not just the ones who have the most [graphics processing units]. But they will be enterprises who have agents they can trust with their data.

Huseni Saboowala, Daxa Inc.