Zero Trust Connectivity for Private AI Apps/Models: Who Can Actually Reach Them?
Zscaler, Monday, August 10th, 2026
Zscaler warns that VPN access to self-hosted AI models grants reachability to entire network segments.
Zscaler argues that while public attention has focused on SaaS AI tools, a quieter trend is enterprises deploying private AI models inside their own data centres - fine-tuned models trained on internal documents, private inference servers at the edge, and agentic pipelines connected to proprietary knowledge bases.
Organisations pursue self-hosting to address data sovereignty, regulatory compliance and sensitive IP protection, and Zscaler agrees this eliminates many third-party data risks. The problem, it contends, is that enterprises secure the model itself while leaving access to it dangerously wide open.