AI Safety Cases

A method in which a developer gathers specific, testable claims about a system into one structured argument that the risks of a given activity are adequately managed, stating its assumptions, uncertainties and residual risk.