This article provides a critical analysis of safety-related assumptions and practices in international nuclear safety regulation, together with an overview of those aspects of safety that have been either signaled or silenced. The data consist of safety reports from the IAEA, OECD NEA and Western European Nuclear Regulators Association (WENRA), as well as from the national stress tests reports of the United Kingdom and Finland. For theoretical tools, we draw on the concept of a dominant co-operative scheme, and on some parts of Luhmann's Theory of Social Systems. The method is content analysis. We argue that the prevailing thoughts on safety by the international nuclear safety organizations, which are based on the field of engineering sciences' understanding of safety, affected what was learned from the Fukushima Daiichi nuclear accident and the way the stress tests were conducted. Moreover, the learning, even though it is still ongoing, has been constrained by some emphasized aspects of safety, which tend to ignore other relevant aspects of safety. Therefore, the stress tests and teachings from Fukushima failed to provide as many improvements in nuclear safety as could have been possible if a broader analysis of the Fukushima nuclear accident was used.