Risk assessment is entirely part of mountaineering, and other risk-encompassing outdoor activities. Decision boundary to go or not go is never a sharp line, and it's always relative to your capacities, your knowledge, and ability to continually assess the risks while practicing.
It is simple to craft the story around an accident in retrospect, but on the field the story is never that simple. If you read accidents reports (like AAC), you will probably never see accidents coming straight from one bad decision, at the beginning of the day, on one particular aspect. It is usual though to see failure about acknowledging changing conditions, multi-factor risks, or even failure to acknowledge high risk/low difficulty transitions by experienced people.
It's also why it is called engagement. Some people are willing to put their life on the line to achieve a moonshot project. But they usually know the risk they are taking.
Llms never took a difficult on field decision, never had to turn gut feeling into calls. If you were experienced enough in both machine learning and mountaineering, you would know a lot of this decision taking is pre-verbal, and is probably very lightly represented in training sets, if at all.