Mount Shasta hikers rescued after using Google Gemini for planning
Sheriff says chatbot advised bringing less food and water, rescue system pays for confident guesses
Images
Image Credits:Christian Richards / EyeEm / Getty Images
techcrunch.com
techcrunch.com
techcrunch.com
techcrunch.com
XDOF's robotic arms manipulate a pen and cap.
techcrunch.com
Three hikers rescued from Mount Shasta after relying on Gemini for planning, sheriff warns against AI-only trip prep, a chatbot’s ‘pack less’ advice meets a real mountain
Three hikers were rescued from California’s Mount Shasta this week after using Google’s Gemini chatbot to plan their trip, according to TechCrunch, citing a report from the Siskiyou County sheriff’s office and coverage by the Chicago Tribune. The men began hiking at 3 a.m., reached the summit at 7 p.m., and then called the sheriff’s office during their descent to ask for directions. They spent the night in Mud Creek Canyon and were rescued the next morning by Forest Service rangers and volunteers.
The sheriff’s office said the hikers had been advised by Gemini to bring less food and water than needed. What was planned as an eight-hour ascent turned into a multiday ordeal, with the usual margin for error compressed by late timing and darkness. In mountain rescue reports, the details that matter are rarely exotic: start time, turnaround time, weather, water, and whether someone has a map and a headlamp. Here, the new variable is not that people sought advice from a tool, but that the tool’s output arrived with the confidence of an itinerary rather than the hedging of a guidebook.
Chatbots are built to answer prompts, not to refuse responsibility. A model can summarize popular routes and list gear, but it cannot see snow conditions, evaluate a user’s fitness, or ask the follow-up questions that experienced rangers use to detect overreach. When a plan fails, the feedback loop is also weak: the cost is borne by the user and the rescue system, while the product that generated the plan typically receives no bill and no liability. The sheriff’s office response points to an older workaround that still scales: contact the local USFS Mount Shasta ranger station before a trip, and treat any generic checklist as a starting point rather than a go-ahead.
The episode also shows how “real-world use” for consumer AI often arrives first as decision support in high-variance environments, where small errors compound. A suggestion to shave weight by cutting water may sound like optimization in a chat window; on a long descent after a late summit, it becomes a constraint that forces riskier choices. Rescue agencies and volunteer teams then absorb the downstream consequences, and the incident becomes a cautionary tale without changing the incentives that produced it.
The hikers got out the next morning with help from rangers and volunteers. The sheriff’s office, in its report, did not recommend a better chatbot—only that people stop relying on one.