Extreme risks are potential events that are very unlikely to occur but which would have a significant impact should they happen. The global financial crisis and its aftermath have demonstrated that risk management cannot afford to stop at the 95th percentile (VaR95) and a holistic risk management framework should include very unlikely, but potentially high impact, events. In essence we are calling for an ‘overweighting’, whether of attention or resources, to be applied to the very low-probability, but bad, outcomes. Why? Because we only live once. Much of finance and economics assume we have infinite lives all running in parallel. This means that the impact of extreme risks can be masked in expected return calculations by frequent positive outcomes. For an individual planning for retirement, a one-in-a-million chance matters because they live in a single universe and face problems in series, not parallel. If that individual gets the one-in-a-million path it could well be terminal – and how they would have fared on the other 999,999 paths is of no useful interest. This paper considers a number of extreme risks. The focus is directed more towards non-economic risks on the grounds that these have received less attention than the more obvious financial and economic risks. However there is one economic risk that we do not think is getting enough attention, namely whether it is possible for society to provide the retirements to which its citizens aspire. The current consensus, among practitioners if not the general population, is that people must save more for their retirement unless they are willing to work until they are very old. While a sensible course of action for an individual, can this be done in aggregate? Is this not a modern version of Keynes’s paradox of thrift? In the context of humanity’s entire history could the recent invention of retirement be seen as an aberration from the norm (of working till you drop, or are supported by family)?