Anthropic shadow is the phenomenon involved when attempts to estimate the probability of a global catastrophe are biased by the fact that they are implicitly conditioning on the existence of human observers.

An event severe enough to destroy all present observers and prevent the emergence of any future observers will necessarily leave no observable traces of its past existence. Such an anthropic effect will bias any attempt to estimate the risk of human extinction based on observed frequencies, causing an underestimation of actual risk.

Milan Ćirković, Anders Sandberg and Nick Bostrom have developed a model to quantify and correct for this effect.^[1]