SAFETY ARCHITECTURE

Code-Red Safety Protocol

Project boundaries, code checks, and approval steps for AI agents.

BOX

Separate workspaces

Agents run work in isolated environments with limits on file and network access.

SEC

Checking untrusted instructions

Checks help identify instructions that try to redirect an agent away from the user’s task.

TLS

Protecting project data

All transit is secured via TLS 1.3, and resting data via AES-256 encryption. Zero customer session data is utilized for model retraining.

CLI

CLI permissions

Aether Code CLI uses access controls and asks for approval before sensitive file operations.

ESC