Google’s AI chatbot, Gemini, has faced targeted “distillation attacks” from hackers aiming to clone its functionality through over 100,000 queries. These attacks, referred to as “model extraction,” involve probing the chatbot for its operational patterns and logic. Google’s Threat Intelligence Group (GTIG) highlighted that such attempts are likely being driven by private companies or researchers looking for a competitive edge. As AI technology evolves, the integration of AI into malicious activities is increasing, making Gemini a potential indicator of future threats against smaller AI tools. Despite having defenses against distillation, large language models remain vulnerable due to their public accessibility. Google views these attacks as intellectual property theft and emphasizes the need for proactive measures to safeguard proprietary information. This evolving landscape poses significant risks, with many firms at risk of similar methodologies as they develop unique AI models influenced by sensitive data.
Source link
