Netskope adds in‑cloud AI guardrails

Netskope has added software that sits inside Google Cloud’s Vertex AI stack to inspect prompts and model replies before they leave an enterprise workflow. (netskope.com) Google Cloud’s own Vertex AI integration for Model Armor already screens requests sent to and responses received from Gemini models, and Netskope is positioning its product as an added control layer for customers building private artificial intelligence apps on Vertex AI. (docs.cloud.google.com) (netskope.com) Netskope says the service can analyze traffic in real time for prompt injection, jailbreak attempts, unsafe content, and data leaks, then apply data loss prevention and threat protection policies before a response is delivered. (netskope.com 1) (netskope.com 2) The basic problem is that large language models take in plain-language instructions and return plain-language answers, which means confidential data can be pasted into a prompt or generated back out in a reply unless something checks both directions. Google documents that Vertex AI and related services have security controls, but some controls vary by feature and setup. (docs.cloud.google.com 1) (docs.cloud.google.com 2) That has turned “guardrails” into a market for runtime filters — software that acts like a checkpoint in the middle of an AI conversation instead of relying only on settings applied before deployment. Netskope’s product page describes that layer as real-time analysis of all traffic, including human and agentic interactions. (netskope.com) Compliance is part of the pitch. Netskope’s Google Cloud materials say joint deployments are aimed at mandates including the General Data Protection Regulation, the Health Insurance Portability and Accountability Act, and other data protection rules, while Google says customers handling protected health information in Vertex AI Search need an appropriate business associate agreement. (community.netskope.com) (docs.cloud.google.com) Netskope has been widening that artificial intelligence security push in 2026. On March 11, the company announced Netskope One AI Security, saying more than 30 Fortune 100 companies use the broader Netskope One platform. (netskope.com) Its documentation also shows the company has been building the pieces around this launch for months, including AI guardrails profiles that let administrators allow or block prompts and responses by category, keyword, or semantic match. (docs.netskope.com) The result is that prompt inspection, response filtering, and policy enforcement are moving from optional add-ons into the normal plumbing of enterprise AI deployments on Google Cloud. (netskope.com) (docs.cloud.google.com)