Safety

Open Settings → Safety.

Platform safety rules

The first table lists platform categories (for example violence, self-harm, hate, harassment). These are locked: every agent on your account is subject to them. You can read the key, label, and instructions the model follows.

Custom safety rules

The second section is Custom Safety Rules. Here you add organisation-specific topics:

• Add Rule opens a form for key (internal identifier), label, and instructions (what the model should do when content matches).
• Edit / Remove apply per row. Keys must be unique (case-insensitive).
• Incomplete rows are filtered out when saving.

Saving updates the account policy and shows a success notification.

Per-agent overrides

Each agent has Agents → Safety where you see the same platform list and can toggle which custom rules apply to that agent. To create or edit custom rules, use Settings → Safety, not only the agent screen.