“A bioterror or bioerror will lead to one million casualties in a single event within a six month period starting no later than Dec 31 02020.” Detailed Terms »

1. Cognitive biases inflate the subjective odds of disaster: If a scenario is imaginable, it seems probable.
2. Moral market forces also distort the odds: Pessimists are seen as serious and responsible, optimists as complacent and naive. Because of 1 & 2, most doomsday predictions are false (thermonuclear war between the US and USSR, Y2K disaster, the population bomb, resource wars, weekly 9/11-scale terrorist attacks, etc.).
3. Bayesian base rate: If an extreme and specific category of event has never before happened, the odds are stacked against it. The only successful past cases of bioterror are the Rajneeshee cult (751 cases, 0 deaths) and the Anthrax mailing (5 deaths). The Soviet-era anthrax bioweapon leak killed another 66.
4. Pool of potential perpetrators: The number of people eager for senseless destruction is small (otherwise we'd have daily bombings, mass shootings, sabotage, etc. in every city, rather than sporadic headline-worthy ones). Multiply by the fraction with the knowledge, skill, intelligence, and discipline to engineer a superbug (as opposed to the bumbling underwear-bomber and Tsarnaev Brother types). The pool may be smaller still: I suspect that being very smart lowers the odds of enjoying mass death and mayhem.
5. Murphy's Law: Multiply further by the fraction who escape detection and accidents (e.g., being stricken by one's own bioweapon), and whose plan succeeds despite the unpredictability of epidemiological processes.
6. Multiply further by the chance that the Black Hat will defeat the White Hats, namely the world's medical and epidemiological networks, who would combat an outbreak with containment, vaccines, antibiotics, etc.
The product is not 0, but I don't think it's high.
Granted, I have the benefit of 15 years of hindsight: I need only three more outbreak-free years. But Sir Martin has the benefit of unforeseen advances in synthetic biology, such as CRISPR/Cas9.
We'd need definitions of "casualty" (death, permanent disability, or just infection?), "bioterror" (state-sponsored? apolitical vengeance/spree killing?) and "bioerror" (I assume a mere lack of preparedness doesn't count). But that shouldn't be too hard.