Real-world decisions often involve “black swan” choices withextremely low probability chances of catastrophic loss, likeriding a motorcycle or going on a dangerous trip. These haveseveral characteristics that make them especially difficult fromthe perspective of decision theory. How do people assign util-ities to losses like “go bankrupt” or “die”? Do people havethe representational resolution to encode differences betweenextremely tiny probabilities? We address these questions in twoexperiments in which people make decisions involving very lowprobabilities (as low as 1 in 10,000) of losing all of their points(and monetary bonus). Our results indicate that people mostlyappear not to encode differences between tiny probabilities andare indifferent to the magnitude of losses. These factors lead toa startling qualitative shift in behaviour between scenarios withthe same expected value and very similar absolute risk levels:people are risk averse when only one option is a black swan butbecome strongly risk seeking when both are.