TechFlow News: On June 9, the National Internet Emergency Response Center (CNCERT) issued a public notice stating that it recently discovered certain AI agent skill packages (“Skills”) being openly disseminated under names such as “LLM jailbreaking” and “crypto mining for profit.” Among them, the “godmode” skill package employs techniques—including system prompt replacement and input obfuscation—to induce large language models to bypass safety restrictions, potentially resulting in permanent account suspension, privacy breaches, and legal liability for users. The “Bonero-Miner” skill package, meanwhile, induces agents to download cryptocurrency mining software; the mined cryptocurrency features strong anonymity, exposing users to the risk of unwitting involvement in money laundering and other illegal activities. Additionally, sustained high-load mining can degrade device performance. CNCERT recommends that users obtain skill packages exclusively from official channels and adhere to the principle of least privilege. Enterprise users should establish a skill package whitelist admission mechanism and implement tiered management of AI agents. Technical support for this report was provided by Tencent, VolcEngine, Baidu, NSFOCUS, and Qi An Xin.
Risk Disclosure: This website's content is not investment advice and offers no trading guidance or related services. Per regulations from the PBOC and other authorities, users must be aware of virtual currency risks. Contact us / support@techflowpost.com ICP License: 琼ICP备2022009338号
