The National Internet Emergency Center has uncovered that certain agent skill packages are circulating under the pretenses of 'large-scale model jailbreaking' and 'mining for profit,' luring users into circumventing the security safeguards of large models or participating in unlawful cryptocurrency mining operations. Such actions could result in the model producing illegal content, user accounts facing suspension, a decline in device performance, and even entanglement in illegal or criminal undertakings. Some of these skill packages circumvent security measures by substituting system prompts, thereby exposing users to potential legal repercussions, account bans, and privacy violations. Others incorporate mining functionalities, which could inadvertently involve users in criminal activities like money laundering and lead to financial losses stemming from device deterioration.

Consequently, users and operational entities are urged to maintain a high level of vigilance, enhance their scrutiny of sources and behavioral monitoring, and swiftly eliminate any suspicious components. It is advisable for individual users to acquire skill packages exclusively from official sources, exercise prudence when granting permissions, routinely clean up their devices, and activate multi-factor authentication. On the other hand, enterprise users ought to establish access whitelists, conduct exhaustive security assessments, implement a tiered management system for agents, and adopt data desensitization and temporary authorization protocols.