Many risks have the property that large numbers of people are exposed and have little or no individual control over the risks they face. Dams, nuclear power plants, and recombinant DNA research have proven controversial not simply because there is vast uncertainty about the true level of risk associated with each, but because a fundamental issue is that in each case social risks can be lessened by spending more on safety: dams can be designed to withstand larger earthquakes, nuclear plants can have additional safety equipment, and DNA research could be done in yet more carefully isolated laboratories. Since each person will have preferences regarding the proper trade-off between increased cost and increased safety, there is little possibility of consensus. More importantly, we show that a voting process for expressing individual preferences can be manipulated and is seriously flawed in the sense that it does not lead to an “efficient” outcome. In addition, we show that virtually all people are unhappy with the safety decision, in the sense that each would prefer either a safer or a cheaper outcome. Thus, making safety decisions that affect a large group of people who will not be able to control the outcome is even more difficult than has been appreciated. We suggest some ways of handling some of the difficulties.