For safety-related AI systems there is sufficient evidence to support a meaningful Safety Management System under JSP 815, collected from the earliest stages where data sets are large or complex. Unique AI safety risks are analysed and written into the overarching safety cases and the software and system risk assessments required by JSP 375 and JSP 376; safety-related risk assessments are carried out early and all risks managed coherently and proportionately through the lifecycle, demonstrating compliance with Def Stan 00-056 and Def Stan 00-055 where appropriate; risk analyses include reasonably predictable behaviour of entities the AI interacts with in benign and adversarial environments and excursions beyond the ODD; and JSP 892 is applied so that consequences of failure reflect direct impacts and wider societal concerns, with risk activities able to keep pace with AI's rate of change.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.