Google warns cyber gangs are extorting companies for stolen AI models and data

Google’s Threat Intelligence Group disclosed that cyber-crime crews are targeting high-value AI assets, stealing them and threatening public disclosure unless a ransom is paid. In one breach, a healthcare firm lost corporate data, drug research and a proprietary AI model; in another, an AI-media company had its source code, prompts, model scripts and secrets exfiltrated. Both attacks were attributed to the UNC6780 group, also called TeamPCP, which has previously carried out extensive open-source supply-chain compromises across PyPI, npm and Docker Hub.

Google’s Mandiant team responded to several such operations in the second quarter of 2026, noting attackers are increasingly embedding agentic AI into multiple stages of their campaigns. The firm cautions that as organizations invest heavily in AI, extortion schemes targeting these assets are likely to rise.