RedTeamKit

RedTeamKit is a web app (with optional CLI) that helps teams stress-test LLM features for common safety and ethics failures before launch. You connect your model endpoint (OpenAI, Anthropic, Azure OpenAI, or self-hosted), pick a risk profile (PII leakage, jailbreaks, self-harm, hate/harassment, medical/legal advice, data exfiltration), and run repeatable test suites. It generates adversarial prompts, evaluates outputs with configurable rubrics, and produces an audit-ready report showing failure rates, examples, and recommended mitigations (policy tweaks, system prompt changes, filters, and monitoring rules). It also supports regression testing in CI so safety doesn’t degrade with each prompt or model update. This is an AI app + traditional app combo: AI generates and judges tests, while the platform manages baselines, evidence, and workflow. It’s pragmatic: it won’t “guarantee safety,” but it will catch many preventable failures cheaply.