It is critical to base scientific risk assessment of genetically engineered organisms (GEOs) on appropriate scientific concepts. A variety of ‘generic safety’ models has now largely been recognized to have been based on outdated scientific thinking. One broad safety argument that is still used is that genetic engineering categorically is nothing but an extension of selective breeding. It is explained here, though, that the mechanisms and potentials of the two can be profoundly different. This does not mean that every GEO is ecologically dangerous; but some types of GEOs may be considerably more risky than what could be produced with selective breeding, especially when an ecologically competent host is supplemented with novel features that may increase its competitiveness. In addition, genetic ‘side effects’ raise food-safety issues; and the possibility that they may sometimes increase ecological competitiveness cannot be ruled out, though this would be quite rare. Field plots have a proper use: to gather particular data that could be used in analysing the risks of commercial releases. But it is not scientific to call a small, confined, field population, isolated from potential competitors, a ‘test or release’ and then conclude that because ‘nothing happened’ the GEO will be safe when commercialized, or indeed that all GEOs will be safe.