Attackers prompted Gemini over 100,000 times while trying to clone it, Google says

In a startling revelation, Google has disclosed that malicious actors have attempted to replicate its Gemini AI chatbot by issuing over 100,000 prompts across various non-English languages. These actions, which Google dubs 'model extraction,' are viewed as a form of intellectual property theft, as the intruders sought to gather data to construct a more affordable imitation of the advanced AI. The findings, released in Google's latest quarterly threat assessment, paint a picture of the tech giant as both a victim and a protector of its innovations. This narrative is common in self-reported evaluations, but it highlights the ongoing challenges posed by competitive entities. The company suspects that many of these attacks originate from private companies and researchers eager to gain an unfair advantage in the AI landscape. Google's history with similar practices adds another layer of complexity to the situation. Earlier this year, it was reported that the Bard team was accused of leveraging outputs from ChatGPT found on ShareGPT, a platform where users share chatbot interactions, to enhance its own model. This led to internal disputes, including a resignation from a senior researcher who expressed concerns over potential violations of OpenAI's terms of service. Despite denying these allegations, Google has since adjusted its data use practices. The company firmly prohibits users from extracting data from its AI models, yet the ongoing incidents reveal a darker side of AI development, where shortcuts like 'distillation' are employed. This technique allows aspiring developers to build new AI models using outputs from previously trained models, circumventing the extensive resources typically required for such advancements. As the battle against these cloning attempts continues, Google remains vigilant, but has chosen not to disclose specific suspects behind the attacks, which are reportedly occurring globally.