PromptBrake

About

PromptBrake enables teams to conduct security assessments on API endpoints powered by large language models using a predefined set of over 60 attack scenarios. Users can link their endpoint, execute scans, and examine results categorized as PASS, WARN, or FAIL, each accompanied by supporting evidence and corrective advice. The tool addresses vulnerabilities such as prompt injection, prompt leakage, data exposure, misuse of tools, and output bypasses, supporting APIs from OpenAI, Claude, Gemini, and numerous custom LLM implementations. API keys remain unretained, and PromptBrake processes findings internally without transmitting scan information to an external language model.