This is a commonly (mis)-used argument, that misses the fact that if the statistics tell us the risk is less than some value (lets say 10^-6), the assumption is that the risk is 10^-6. Often, statistical tests have the power to detect very rare effects, but not effects that are even rarer. So less than 10^-6 does not mean the risk is one in a million, it means we do not know how low the risk is (could be 10^-23), but we know that it is no higher than 10^-6.