Gemini Underestimated Supplies: AI as a Mountain Guide Almost Killed Three People

Mount Shasta is unforgiving. When a trio of young men set out for the summit at 3 AM, they ignored a basic rule: if you aren't at the summit by noon, turn back. They didn't reach the top until 7 PM. Their subsequent attempt to descend in the dark ended with an emergency call for help and spending the night in Mud Creek Canyon. They had to be rescued by rangers and volunteers in the morning.

Hallucinations with fatal consequences

According to a report from the Siskiyou County Sheriff's Office, artificial intelligence played a key role in their planning. The hikers relied on the Gemini model, which recommended an amount of food and water completely inadequate for the difficulty of the ascent. What was supposed to be an eight hour hike turned into a multi day struggle for survival without enough supplies. The model failed to properly assess the physical demands or the safety margins necessary for alpine terrain.

A false sense of security from the screen

People tend to trust technology more than their own judgment. When your smartphone tells you a liter of water is enough, it's easier to ignore the warning signs. This phenomenon, known as automation bias, combined with the persuasiveness of language models, becomes a deadly cocktail.

A chatbot is not a local expert

The case highlights the critical risk of delegating physical safety to language models. They are trained to provide confident and coherent answers, even when they have no concept of real world conditions, weather changes, or the specifics of a given terrain. Rescuers therefore warn people to always contact local stations and never rely solely on AI when planning trips into nature.

The reaction of tech giants will decide

Google has not yet officially commented on the incident, but this case will likely lead to stricter rules and warnings within the Gemini app when a query involves planning trips to dangerous areas. The real question is how quickly companies can implement safety filters to prevent such advice.

Lilith's verdict

An LLM doesn't know what thirst feels like because it has no body. Before taking survival advice from a chatbot, remember that it doesn't care if you freeze to death on a mountain.