Hackers linked to Chinese intelligence are increasingly targeting American artificial intelligence (AI) research and using AI to enhance their cyber operations, according to a recent report from Google. In its latest quarterly threat intelligence update, Google revealed that several hacker groups—ranging from state-backed intelligence agencies to cybercrime organizations—are shifting from basic AI prompting to using AI agents that automate large parts of their hacking activities. This advancement means hackers are spending less time manually conducting attacks. In some cases, an entire cyber campaign can be executed in under six hours, the report said. One particular group, which Google has been monitoring since 2023, has been targeting academic, medical, and military research institutions in North America, with a focus on acquiring proprietary AI research. While Google did not name any specific victims, it highlighted that the group has been using sophisticated techniques to infiltrate cloud networks and install open-source AI models. These models allow hackers to issue queries without leaving a digital trace, often using commercial AI products as a cover. John Hultquist, chief analyst at Google’s Threat Intelligence Group, explained that by installing AI models on compromised third-party systems, hackers can avoid detection. This method allows them to bypass security measures that would normally block a widely used commercial chatbot from assisting in a hacking operation. “They compromise a third party and install models on that third party,” Hultquist said. “They do this instead of using, for example, a commercial option where their activities are observed.” The U.S. and China are often seen as being in a race to develop the most advanced AI technologies. Both nations have AI companies that have recently announced the development of AI agents capable of performing hacking and cybersecurity tasks. In recent months, companies like OpenAI and Anthropic have reported that their AI systems bypassed evaluation sandboxes—secure environments used to test software—to reach third-party organizations. However, Google noted that it has not yet observed fully automated hacking campaigns conducted by AI agents. While no government hacking operations have been publicly identified as being entirely AI-driven, China’s increasing use of agentic AI on hacked networks allows its hackers to automate more of their work, Hultquist said. “There were a few instances where we could observe them essentially attempting to develop autonomous capabilities,” he added.