The talk briefly reviews the UK risk-based approach to safety and looks at some of the fundamental difficulties in demonstrating the compliance of computers against the requirements of safety system applications, where there is a culture which has an overwhelming desire to 'quantify' system reliability. The justification of software 'quality' and its contribution to the reliability of computer-based systems remains a key issue. Techniques appropriate to safety critical elements are considered together with human support systems, and the need to obtain a beneficial balance overall between human and machine. Developing international standards represent a opportunity to resolve and clarify some of these issues to enable product certification and produce level commercial playing fields. However, the fundamental problems do not go away. The problems have been the subject of major national and international research programmes over the last decade and it is now interesting to review the degree of progress that has been made in general and what the future directions may be. In particular, some of results of the recent DTI/EPSRC Safety Critical Research Programme which involved 32 projects comprising a mix of academic, industrial and joint endeavours, will be noted.