Almost two decades' experience in building expert systems has shown that strong development methodologies are desirable if we are to ensure that systems are effective. Logic programming techniques combined with software engineering methods offer a potent discipline for ensuring that software is sound. However, if we are to achieve trust in any medical technology it is not enough to demonstrate soundness. We must also show, so far as is possible, that it is safe. Alongside the pursuit of rigorous software development techniques, therefore, we need to investigate general design and development methods that minimise the occurrence of hazards. We can adopt some general techniques from existing safety engineering practices, but AI also supplies some important new ideas for improving safety.