Three Hikers Get Lost on Mount Shasta After Gemini Planned Their Route

Three young men from Roseville, California set out at 3 a.m. on Sunday, August 30, to climb Mount Shasta, expecting an eight-hour trip. They had consulted Google's Gemini chatbot beforehand on the route, pace, and supply list. The ascent alone took sixteen hours, the descent turned into a night spent in a canyon without proper gear, and the whole ordeal ended with a mountain rescue operation.

Advice That Failed Them

According to the account the men gave rescuers, they based their preparations almost entirely on Gemini's guidance. The chatbot underestimated the climb at eight hours, even though experienced climbers typically need well over ten hours on this route. It also advised packing simple carbohydrates instead of fats, reasoning that fat supposedly takes too long to digest - as a result, the group brought far too little food and water, with no proper shelter or warm clothing for a night in the mountains.

The Siskiyou County Sheriff's Office put it bluntly in an official statement after the rescue. The miscalculation proved critical: what was planned as an eight-hour climb turned into a multi-stage ordeal for which the food and water they'd packed simply weren't enough.

This was a critical error, as Gemini advised them to bring far less food and water than the group needed - Siskiyou County Sheriff's Office

A Night in the Canyon

The group reached the summit at 7 p.m., many hours after the early-afternoon turnaround time experienced mountain guides recommend to ensure a descent before dark. On the way down, already in darkness, one of the men twisted his knee at an elevation of about 3,500 meters. An hour into the descent, the phone running the AllTrails navigation app died, and the group strayed off route into the canyon of Mud Creek on the mountain's south side. After midnight, the hikers stopped at around 3,350 meters and decided to wait out the night in the open, having already called for help by phone.

The Rescue

On the morning of August 31, U.S. Forest Service climbing rangers reached the exhausted, hungry and dehydrated hikers along with a search-and-rescue team from the Siskiyou County Sheriff's Office. Rescuers treated the injured knee and gave the men food and water. Weather conditions ruled out a helicopter evacuation, so walking all three men down took another nine hours. The hikers themselves later admitted they had relied on artificial intelligence more than on their own judgment.

We relied too much on AI instead of our own critical thinking - Mount Shasta hikers

A Warning From Authorities

In its post-rescue statement, the sheriff's office urged people never to rely solely on artificial intelligence when planning high-altitude expeditions, and recommended calling the Mount Shasta ranger station beforehand to get verified, up-to-date information on trail conditions. This is not the first case in recent months where an AI chatbot has misled hikers about the difficulty or duration of a mountain route, resulting in a call to emergency services.

Mount Shasta is a dormant stratovolcano in Northern California, one of the tallest peaks in the Cascades, popular with climbers from across the United States precisely because its summer summit route is relatively straightforward, if physically demanding. It's also a route where sudden weather shifts and fast-falling darkness regularly catch poorly prepared groups off guard, regardless of where they got their information beforehand.

For Polish readers, the story from California sounds familiar - Polish mountain rescuers have voiced similar warnings for years, reminding hikers that navigation apps and AI-generated advice are no substitute for experience, a backup paper map, or checking the forecast directly with local rescue services. More and more people planning routes in the Tatras or the Beskids are turning to chatbots, and this season TOPR and GOPR (Poland's Tatra and Beskid volunteer mountain rescue services) repeatedly had to intervene after hikers trusted app-suggested shortcuts that led into difficult climbing terrain.

The Mount Shasta case feeds into a broader debate over how responsible language model makers are for advice that affects users' physical safety. Gemini, like other large models, can generate confident-sounding but wrong estimates of time, distance or required supplies, because it draws on averaged data from the web rather than a real assessment of a specific group's fitness, experience or the current weather conditions. Google has not publicly commented on this specific incident so far.

The case didn't result in fines or criminal charges, but local rescue services stress that operations like this cost time, tie up limited mountain rescue resources, and could end tragically in worse weather. A growing number of similar incidents in the US and Europe suggests the problem won't be limited to a single case until popular chatbots start flagging the uncertainty in their recommendations more clearly for activities that carry real risk to health and life.