Recently, JD.com officially launched its open-source large model security framework - JoySafety. This framework has undergone multiple internal tests and has been widely applied in many business scenarios of JD.com, including AI shopping guides, logistics customer service, sales assistants, medical consultations, merchant workstations, legal consultations, and security Q&A. These scenarios have driven the creation of JoySafety, aiming to provide enterprises with a mature, reliable, and free security protection solution.

The JoySafety framework supports over 100 million calls per day, with an attack interception rate of over 95%, ensuring that users can enjoy efficient services while their data security and privacy are effectively protected. The design philosophy of the framework is to balance user experience and security, allowing various services to run smoothly without being affected by potential threats.

image.png

The core technical highlights of this framework lie in its multi-turn conversation intelligent identification capabilities. JoySafety adopts an innovative Free-Taxi asynchronous output mode, which significantly improves the system's detection response speed, ensuring a smooth user experience during use. In addition, JoySafety introduces a multi-dimensional response strategy matrix, which can take different response strategies based on the situation. For example, when high-risk content is detected, the system will immediately block the relevant information; when facing sensitive questions, the red line knowledge base will provide appropriate responses; meanwhile, the system also has intelligent correction functions that can automatically guide users towards positive information.

With the rapid development of AI technology, security issues are becoming increasingly prominent. The launch of JoySafety undoubtedly provides an important security barrier for enterprises using AI technology. In the future, JD.com will continue to focus on technological innovation, safeguarding user safety and trust.