Ben Pace comments on Catastrophic sabotage as a major threat model for human-level AI systems