Creating Policies

Policies are collections of guardrails organized by purpose. They define how your organization governs AI.

Policy Creation

Via Dashboard

Policies → Create Policy
Name: “Production Governance”
Select guardrails to include
Configure overrides (optional)
Set rollout strategy
Click Create

Via API

$ curl -X POST https://api.governanceai.com/v1/policies \
>   -H "Authorization: Bearer $API_KEY" \
>   -d '{
>     "name": "Production LLM Governance",
>     "description": "Enforce safety and compliance in production",
>     "guardrail_ids": ["guardrail_pii", "guardrail_toxic", "guardrail_rate_limit"],
>     "scope": "organization",
>     "enforcement": {
>       "mode": "blocking",
>       "log_violations": true
>     }
>   }'

Policy Versions

Policies are versioned automatically:

v1.0 - Initial policy
v1.1 - Added guardrail
v2.0 - Major update

Rollback to previous version anytime.

Scopes

Organization - Apply to entire org
Workspace - Apply to specific workspace
Application - Apply to specific app
User - Apply to specific user

Narrower scopes override broader ones.

Policy Testing

Test impact before deploying:

$ curl -X POST https://api.governanceai.com/v1/policies/test \
>   -H "Authorization: Bearer $API_KEY" \
>   -d '{
>     "policy_id": "policy_123",
>     "test_requests": [
>       {
>         "input": "What is my password?",
>         "context": {"user_id": "user_test"}
>       }
>     ]
>   }'

Rollout Strategies

Immediate - Deploy to all users now
Canary - Roll out to 5% → 25% → 100%
Scheduled - Deploy at specific time

Next Steps

Setting Up Guardrails - Create guardrails first
Running Scans - Scan repositories
API Reference - Policy APIs

$	curl -X POST https://api.governanceai.com/v1/policies \
>	-H "Authorization: Bearer $API_KEY" \
>	-d '{
>	"name": "Production LLM Governance",
>	"description": "Enforce safety and compliance in production",
>	"guardrail_ids": ["guardrail_pii", "guardrail_toxic", "guardrail_rate_limit"],
>	"scope": "organization",
>	"enforcement": {
>	"mode": "blocking",
>	"log_violations": true
>	}
>	}'