It's not bullshit, and it really does apply to anything. Observe people who had only theoretical awareness of some issue (e.g. kidnappings, bad treatment of animals) who shrug off statistics, but if they read a detailed story about a particular example, they're suddenly super upset about it.
Specifically in the context of this bet, it's semi-bullshit, because we have no number of how inflated the estimate of danger is.
You can't really say "Rees must be having a cognitive bias, and his estimate of the bioterror/bioerror has to be grossly overstated". Rees is not a population. the Law of large doesn't apply. He might have been too optimistic. Even if he is overly pessimistic, maybe the inflation of risk isn't that significant. Or it is significant, but the risk is so great, that even after subtracting the inflation, it is still dangerous.
The point is, applying a statistical heuristic without supplying the actual parameters is meaningless. The heuristic itself is not bullshit. The application in this case certainly is.
The bias itself is probably a cousin of availability heuristic - https://en.wikipedia.org/wiki/Availability_heuristic.