Large Language Models At Risk For ‘model Leeching’ Attacks

Researchers from Mindgard and Lancaster University have demonstrated that chunks of Large Language Models (LLMs) used in AI chatbots can be copied in less than a week for as little as $50, and the information gained can be used to launch targeted attacks. The attack, termed “model leeching,” works by talking to LLMs in such a way so that the LLMs elicit insightful information giving away how the model works. The research team focused their study on ChatGPT-3.5-Turbo and were able to use this knowledge to create their own copy model, which was 100 times smaller but replicated key aspects of the LLM. They were then able to use the knowledge gleaned from their model to attack vulnerabilities in ChatGPT with an 11 percent increased success rate.