Google Reports Over 100,000 Attempts to Clone Gemini AI

Google has reported that its Gemini AI chatbot has been targeted by over 100,000 prompts from private sector entities attempting to extract its proprietary reasoning processes. The company describes these "model extraction" attacks as a form of intellectual property theft.

In a recent report, Google's Threat Intelligence Group (GTIG) detailed how these entities have been submitting carefully crafted prompts designed to reveal Gemini's internal logic. This technique, known as "distillation attacks," involves repeated questioning intended to uncover the patterns and logic that underpin the AI's responses. The attackers aim to use the extracted information to build or enhance their own AI systems.

Google has taken steps to mitigate these attacks by disabling associated accounts and strengthening security controls within the Gemini models. The company also warns other AI developers to be vigilant against similar model extraction attempts, emphasizing the potential risks to proprietary AI technologies.