ASRA-Q: AI Security Risk Assessment by Selective Questions

In this paper, we propose a new framework for security risk assessment. To conduct analysis efficiently, it is necessary developers to assess the risks of machine learning based system (MLS) by themselves, but existing technologies cannot be used such purpose. Using proposed framework, MLS can MLSs themselves. Our consists two phases. preparation phase, expert extracts conditions adversarial attacks each attack method and makes an tree using extracted conditions. addition, they prepare yes/no questions corresponding assessment just answer questions, results are shown. We asked some evaluate our proposal implementing framework. As result, found vulnerabilities in chose analyze. received positive comments from them as questionnaire.