Please report vulnerabilities privately via GitHub: Security → Report a vulnerability on this repository (private vulnerability reporting is enabled). Please don't open a public issue for a security report. You'll get a response within a few days; confirmed issues ship as a patch release, credited to you unless you prefer otherwise.

The latest release receives security fixes.

• Delegated-permissions ceiling — OpenGateSP talks to SharePoint as you (or as your app-only certificate identity); it can never exceed the permissions of the signed-in identity.
• AI keys stay local — assistant API keys are encrypted with Windows DPAPI (current user, this machine) in %APPDATA%\OpenGateSP\aiconfig.json; they are never stored in plain text and never sent anywhere except the provider you chose.
• Tenant content is untrusted input to the AI assistant — write actions are off by default; when enabled, the model only sees schema-declared tool arguments, every write runs as a preview first, and the apply call is honored only after you reply to the preview in a later message. These are code-enforced guarantees, not model behavior (see docs/13-ai-assistant.md).
• Write operations preview first everywhere — the GUI wizard, the MCP tools, and the cmdlets default to dry-run and require explicit opt-in (execute: true / -Force) to change anything.