As AI agents take on enterprise tasks, companies face a new battle over access and control
SiliconANGLE Ryan Stevens ● Covered by 3 sources
AI agents are moving into real enterprise work, and companies are racing to lock them down. The twist: the winners may be the ones that can give agents access without letting them roam.
Based on reporting by SiliconANGLE, Ryan Stevens — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Enterprise AI has moved beyond pilots, and that’s created a messy new problem. Agents can write code, process claims and run workflows, but they need models, tools and data to do any of it. Few companies want software with that much autonomy loose inside their systems, so platform teams are suddenly back in the middle of the conversation.
Purnima Padmanabhan, general manager of Broadcom’s Tanzu Division, said the old cloud-native playbook is coming back in a new form. Orchestration, telemetry, observability and role-based access control are now being used to manage fleets of agents. Her point is simple: if agentic apps behave like microservices, then the platform is where control belongs.
Broadcom’s approach is a runtime, not just a toolkit. Tanzu Platform Agent Foundations now sits inside VMware Private AI Cloud, giving agents a sandboxed environment where every model, tool and dataset has to be explicitly bound. Curated data products handle chunking, vectorization and access control, so agents can work with prepared data without reaching straight into source systems.
Padmanabhan argued that the trick is not to starve an agent of access, but to give it access on strict terms. Identity, role-based access control and credential management all matter, and so does the ability to keep everything on one platform. Broadcom is also scanning its commercial repositories and open source with Mythos, while extending security work in Spring and Java into Python and Node.js.
That security push is not theoretical. Padmanabhan said Broadcom is finding vulnerabilities, including ones that could “completely bring down an enterprise.” She also pointed to June’s patch release as the largest in 23 years of Spring stewardship. The message to CIOs, in her view, is that this is both a revenue opportunity and a competitive advantage, and hesitation could become its own disadvantage.
My take — AI-written commentary, not fact-checked reporting
This is the part of enterprise AI people keep trying to skip: agents are not magic, they’re just software with a fancier sales deck and a bigger permission set. The real battleground is control, not prompts, and that’s exactly why platform teams are having a comeback. Open or closed models matter less than whether anyone can explain who gave the agent access to what, and why.
Read more about this at: SiliconANGLE
