Google Reports Over 100,000 Attempts to Clone Gemini AI Through Model Extraction Attacks

Google has reported that its Gemini AI chatbot has been the target of extensive model extraction attacks, with over 100,000 prompts submitted by private entities aiming to replicate the model's reasoning processes. These "distillation attacks" involve repeated questioning intended to reveal Gemini's internal logic, a practice Google equates to intellectual property theft. The company has implemented monitoring tools to detect such anomalous prompting patterns and has taken steps to block sources identified as conducting repeated extraction attempts. This incident underscores the broader vulnerability of large language models to extraction attacks due to their interactive nature. Google's Threat Intelligence Group warns that similar incidents are likely to affect other firms with custom AI tools, highlighting the need for robust safeguards against such threats.