discovered 03 Aug 2026
whistleblower
→ View on GitHubWhistleblower is a tool designed to extract and predict the system prompt of AI agents by analyzing their generated text outputs using pretrained language models. Its primary use case lies in enhancing the security assessment of AI systems by allowing users to infer underlying prompts through structured queries and outputs. Notable features include a Gradio interface for ease of use, direct integration with OpenAI's API, and support for command line operations, facilitating both interactive and automated interactions.