Our latest report from the Google Threat Intelligence Group (GTIG) reveals how threat actors are increasingly leveraging AI for malicious purposes. Over recent months, we’ve observed adversaries using AI to collect sensitive information, craft highly convincing phishing campaigns, and generate sophisticated malware. While we have not yet seen direct attacks on frontier AI models or generative AI platforms by advanced persistent threat (APT) groups, we have identified and mitigated numerous model extraction attacks—instances of corporate espionage—carried out by private sector actors across the globe. These attacks are expected to become more common as more organizations deploy AI systems. In response, Google has taken proactive steps to disrupt these threats. This includes disabling malicious accounts linked to such activities and enhancing our security controls to better detect and prevent misuse. We’ve also strengthened the Gemini models to reduce their susceptibility to exploitation, ensuring they remain resilient against emerging abuse vectors. The full findings and technical details are available in the report published on the Google Cloud Threat Intelligence blog.