Three hikers rescued on Mount Shasta after Gemini-planned climb ran short on supplies

What happened

Three young men had to be rescued from Mount Shasta in Northern California after planning their climb with Google's Gemini chatbot, TechCrunch reports. The account, which TechCrunch attributes to the Chicago Tribune and to a report from the Siskiyou County sheriff's office, describes a trip that unraveled step by step.

The group set off at 3 a.m. Climbers on the mountain are told to turn around if they have not reached the summit by noon; according to the sheriff's report, these three topped out at 7 p.m., seven hours past that turnaround point. They then tried to descend in the dark, at some stage phoned the sheriff's office to ask for directions, and ended up spending the night in Mud Creek Canyon. Forest Service rangers and volunteers brought them out the following morning.

What the sheriff's office said

The detail drawing the most attention concerns supplies. According to the sheriff's office, Gemini advised the group to bring substantially less food and water than they actually needed — a shortage that became acute when the roughly eight-hour ascent they had planned stretched into a much longer ordeal.

TechCrunch adds a caveat worth keeping in view: it is unclear how much of the blame belongs to the chatbot. The hikers' own choices — pressing on long past the noon turnaround and then attempting a night descent — clearly contributed to the outcome. The supply recommendation, however, is the part the sheriff's office tied directly to the AI tool.

The office's advice was straightforward: call the local US Forest Service Mount Shasta ranger station before setting out to get accurate, current information, and never rely solely on AI when planning a trip.

Confident output, no local knowledge

The failure mode here is a familiar one for large language models. Chatbots produce fluent, confident-sounding answers, but they are not authoritative sources and have no direct knowledge of a specific mountain on a specific day. Local safety conventions — such as a hard turnaround time — may not make it into a generated itinerary, and estimates for essentials like water needs on a long climb can land low without any signal to the user that the number is soft.

In mountain travel, margins are the entire point: extra food, extra water, buffer daylight. A plan built on the assumption that everything goes right becomes dangerous precisely on the day something goes wrong, which is what happened on Shasta.

Why it matters

This incident is one of the clearer public examples of AI-assisted planning producing physical-world consequences. Most AI error stories involve hallucinated facts, broken code or bad summaries; this one ended with a call to a sheriff's office, a night in a canyon and a rescue by rangers and volunteers.

It also marks a shift in how officials talk about these tools. A county law enforcement agency has now explicitly told the public not to treat AI as a self-sufficient trip planner — guidance that implicitly acknowledges how routinely chatbots are being used for exactly that purpose.

The practical lesson is less "never use AI" and more "verify, then add margin." A chatbot can be a reasonable starting point for research and organization, but its output needs checking against people and sources with genuine local knowledge, and any high-stakes plan should include the extra food, water and time that an optimistic generated itinerary tends to leave out.