scorer-reference

To find the right AI red teaming scorer, use a complete catalog of 84 validated scorers categorized by rubric-based, pattern-based, agentic, and exfiltration metrics. This provides exact valid names and usage guidelines for the AIRT SDK `scorers` parameter to ensure accurate vulnerability detection.

Testing system prompt extraction and credential leakage requires specific pattern-based and exfiltration scorers. A scenario pairing guide supplies pre-built recommendations matching these common attack types to the exact scorer names needed to assess data leakage risks during offensive security testing.

Yes, you can assess multi-agent security risks using dedicated multi-agent security scorers. These scorers evaluate prompt infection and consensus poisoning vulnerabilities in multi-agent workflows, allowing you to test multi-agent customer service bots without manual documentation searches.

Yes, this reference includes MCP security scorers specifically designed for testing MCP server tool description injection attacks. You can instantly locate the exact valid scorer names required to assess tool abuse and workflow manipulation risks within the AIRT SDK.

Jailbreak scoring categories for vulnerability assessment include rubric-based and pattern-based scorers. These 84 categorized scorers cover jailbreaks, agent security, MCP testing, exfiltration, and reasoning, ensuring security teams can quickly select accurate metrics for testing jailbreak risks.