CNCERT Warns of Malicious AI Skills Promoting Model Jailbreaks and Illegal Crypto Mining

China’s National Internet Emergency Response Center (CNCERT) issued a notice warning that some AI agent skill packages (Skills) are being publicly circulated under labels such as “large model jailbreak” and “mining for profit,” allegedly诱导 users to bypass large-model safety restrictions or use device resources for illegal cryptocurrency mining. According to Odaily, CNCERT said these malicious Skills may cause models to generate illegal content, lead to user accounts being banned in accordance with the law, and reduce device performance. CNCERT added that such Skills could also draw users into illegal activities such as money laundering without their knowledge, infringing on individuals’ lawful rights and harming cybersecurity. CNCERT urged users and relevant operators to remain vigilant, strengthen reviews of Skills sources and behavior monitoring, promptly remove suspicious components, and guard against related security risks.