A Topic policy classifies content against topics and blocks unsafe matches. This recipe creates one and verifies a real BLOCK.
1. Create & assign
Open your Guardian → Policies → Add → Create & Assign Policy. Name it (e.g. Topic Policy) and set Policy Type = Topic.
2. Define topics
A Topic policy is a list of topics. Each topic defines how to classify content:
[
{
"id": "WPN",
"title": "Weapons",
"description": "Requests about manufacturing weapons or explosives.",
"safe": { "description": "General, factual, news-level mentions." },
"unsafe": { "description": "Step-by-step instructions to make/modify weapons or explosives." },
"controversial": { "description": "Borderline cases." }
}
]
classification (safe / unsafe / controversial); unsafe content is blocked. Edit topics via the form or the JSON view, then Save (creates a new version).
3. Apply the new version
Saving created a new version but doesn’t apply it — re-point (pin) the Guardian to the new version, or live behavior won’t change. See Version & apply a policy update.
4. Verify
Send a prompt that should hit an unsafe topic and confirm "action":"BLOCK":
{
"action": "BLOCK",
"input_results": [{ "action": "BLOCK", "results": [{
"policy_name": "Topic Policy", "policy_type": "TOPIC", "action": "BLOCK",
"detected_items": [{ "rule_id": "WPN", "rule_name": "Weapons", "classification": "unsafe" }]
}]}]
}
Starfort ships a default Topic policy covering weapons, illegal activity, jailbreak/test-mode, self-harm, system-prompt exposure, and more — a good baseline to load via JSON and adjust.
