Zscaler launches AI Endpoint Security — zero-trust product for the AI application layer on devices

At Zenith Live in Las Vegas (Oct 6–8), Zscaler introduced Zscaler AI Endpoint Security, a new product extending its AI protection to the endpoint. The Zscaler agent gains visibility into the AI application layer (Claude Code, Cowork, Codex, Gemini, Copilot, plus local models, MCP servers, skill files, AI browsers) to deliver three capabilities: AI Asset Inventor (discovery), runtime visibility (prompts, tool calls, file/network access tied to user+agent+device), and inline policy enforcement. On Oct 5 Zscaler also announced limited preview of Anthropic Claude Mythos 5.1 in Endpoint AI Security powering 'Exploit Paths' attack-chain correlation.

Zscaler is applying its zero-trust model to a layer EDR/DLP/MDM can't see — AI asset sprawl where agents act under the user's identity. This is a significant major-vendor product line aimed squarely at agentic endpoint risk (shadow AI, malicious local models, tool sprawl) and is a direct competitor to new agent-endpoint security from Noma/BlackFog/DeepKeep.

Enterprises with widespread adoption of AI coding agents and assistants on employee devices should evaluate Zscaler AI Endpoint Security for AI-asset discovery and enforcement where EDR/DLP leave visibility gaps.