Three hikers learned the hard way this week that Google’s Gemini is not a backcountry guide. The trio used the AI chatbot to plan a Mount Shasta expedition and ended up stranded overnight in Mud Creek Canyon before Forest Service rangers and volunteers pulled them out the next morning, according to the Siskiyou County sheriff’s office.
Their itinerary was doomed from the start. The group set off at 3am, hit the summit at 7pm — seven hours past the noon turnaround time that every responsible climber knows — then tried to descend in total darkness. They eventually called the sheriff’s office for directions and spent the night in the canyon. The sheriff’s office said Gemini advised them to bring far less food and water than they required, which turned a planned 8-hour ascent into a multiday ordeal.
What’s striking here isn’t that an AI gave bad advice. It’s that nobody in the group questioned it. These weren’t novices ignoring basic safety protocols by accident; they followed a confident-sounding recommendation from a model that has no concept of exposure, weather windows, or what happens when your water runs out at 10,000 feet. Gemini doesn’t know what it doesn’t know, and that’s precisely the problem when people treat it like a subject-matter expert.
The sheriff’s office put it bluntly: call the local USFS ranger station before your trip and never rely solely on AI for trip planning. That advice will age well. This is the same pattern we saw with early GPS systems routing drivers onto unpaved roads — except the stakes here are hypothermia and death, not a blown tire. The question isn’t whether AI will get better at this. It’s whether people will still be alive to use the improved version.
