Show HN: Capture the Flag game where LLMs are the only players

Category: infrastructure

Tags: capture-the-flag, llm, cybersecurity, docker, self-play, fine-tuning, quantization

Score: 7.5/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 7)

A research platform where multiple LLMs compete in a capture-the-flag game within isolated Docker containers, attacking and defending vulnerable targets. It uniquely combines AI-vs-AI cybersecurity simulation with a custom bot training pipeline that uses LoRA fine-tuning and self-play on game replays, enabling comparative model studies and iterative model improvement.

Target audience: AI researchers, security researchers

Repository: https://github.com/Megapixel99/capture-the-flag · MIT

View on Hacker News