In recent years, the rapid development of artificial intelligence, with the home, finance, transportation, medical and other fields of deep integration, so that people's lives more convenient. But at the same time, if the system based on artificial intelligence has risks, it will bring more serious consequences. How to ensure that artificial intelligence will not be easily controlled, influenced or cheated in different application scenarios has become an important research topic for practitioners.
September 25, Tencent officially released the industry's first AI security attack matrix. This is a highly practical AI security technology guidance framework, the first comprehensive combing of academic and industrial cutting-edge AI security research, From the perspective of the attacker, the attack process and technology realization means in the development and deployment of AI technology are listed.
At present, the full version of risk matrix can be downloaded for free on Tencent AI Lab official website.
Full text download address: https://share.weiyun.com/8InYhaYZ
AI security threat risk matrix
The matrix is compiled jointly by Tencent AI lab and Vermilion Bird Labs, and draws on the ATT of open source security research rack with high maturity and practical significance in the field of network attack and defense
The threat risk matrix of AI security emphasizes the real scene, and classifies attack technology intuitively according to the maturity, research and potential threat. According to Tencent AI Lab, the core difficulty of matrix compilation is how to select and sort out the analysis angle of AI system security problems. As an application that works with other software and hardware, the angle of AI system security analysis is not completely consistent with that of traditional Internet products. After full investigation, the team finally chose to start from the perspective of AI R & D deployment life cycle, summed up the security risks faced by AI system in different stages, and looked at Ai's own security from a global perspective.
Under the guidance of this practical idea, the matrix can be used as convenient as a dictionary. According to the basic situation of AI deployment and operation, researchers and developers can check the possible security problems against the risk matrix, and reduce the known security risks according to the recommended defense suggestions.
n addition to focusing on four basic research areas, such as machine learning, computer vision, speech recognition and natural language processing, tencent AI Lab is also continuously focusing on security research in the AI field, helping to design and deploy credible AI systems. Tencent Vermilion Bird Laboratory focuses on actual combat attack technology research and AI security technology research to attack and defense, guarding Tencent business and user security. Before that, Vermilion Bird laboratory had simulated the hacker attack path in actual combat, and directly controlled the neurons of the AI model, which was the model
Yang Yong, vice president of Tencent cloud and head of Tencent Security Platform Department, said that with the deepening of AI application in the industry, there will be more practical attack methods and means at all levels. Tencent will continue to study the AI security field and update the risk matrix from time to time to keep pace with the latest achievements in the industry.