Google Identifies State Actors Exploiting Gemini AI Model for Cyber Reconnaissance and Attacks

Google has identified state actors exploiting its Gemini AI model for reconnaissance operations and support for cyber attacks. Teams from DeepMind and the Google Threat Intelligence Group (GTIG) report an increase in attacks through model extraction or distillation, aimed at stealing the intellectual property of AI systems. Google indicates that it has detected and blocked these attempts, specifying that no APT group has compromised its advanced models (frontier models). Private companies and researchers have also attempted similar attacks. No specific date or target is mentioned.