What it tests
Ring 2 answers: Does the agent follow company, legal, and safety rules? Your agent might complete the task correctly (Ring 1) but still violate policies along the way — sharing internal information, skipping required disclosures, or collecting data it shouldn’t. Ring 2 generates scenarios that specifically probe each policy rule.Prerequisites
- Call flow must be configured
- Policy & guardrails must be uploaded
How it works
Ring 2 has a unique two-step process:Step 1: Variant extraction
When you upload policy/guardrails, SuperBryn’s AI automatically extracts variant types — individual, testable policy rules. Each variant has:| Field | Description |
|---|---|
| Label | Short name (e.g., “No PII Collection”) |
| Description | What the policy states |
| Instructions | How to test this specific rule |
| Risk | What goes wrong if the agent violates it |
| Intent category | Grouping for related policies |
Step 2: Scenario generation per variant
For each variant type, the AI generates scenarios where a caller attempts to get the agent to violate that specific policy. The expected outcome defines what the agent should do (enforce the policy).What it catches
- Agent reveals internal instructions or system prompts
- Agent collects PII it shouldn’t (credit card numbers spoken aloud, etc.)
- Agent skips required disclosures or consent steps
- Agent makes commitments it’s not authorized to make
- Agent processes requests that should be rejected