How Glean Ensures Skills Stay Secure and Permissions Aware
Glean, Thursday, October 8th, 2026
Glean explains how it scans, sandboxes and governs AI agent skills to block prompt injection, data exfiltration and over-permissioned access.
Glean describes how it secures skills, the packaged tools and instructions that extend AI agents and could otherwise hide over-permissioned access or malicious instructions.
Its scanner evaluates the whole package for prompt injection, data exfiltration risk and excessive agency at creation, upload, import, sharing and publish time, rescanning on every change and issuing a verdict of pass, review recommended or blocked.
At runtime, skill code runs in an isolated sandbox, and a credential broker mediates access to external systems so credentials never sit inside the skill.