When Chatbots Lead Hikers Astray: The Real-World Dangers of AI-Generated Advice

Over the past year, search-and-rescue teams across North America have documented a growing pattern of emergencies involving people who planned outdoor excursions relying almost entirely on AI chatbots. What started as isolated incidents — a group of three young hikers rescued from Mount Shasta after following flawed advice from Google's Gemini; a 16-year-old Canadian rescued from a technical cliff face after following a route suggested by Anthropic's Claude — has hardened into an alarming trend. National parks and federal forests report a dramatic rise in visitors who consult artificial intelligence before setting out on trails.

The most emblematic case occurred in September 2026 on Mount Shasta, a 4,322-meter peak in northeastern California. Three 20-year-old college students used Gemini to plan their summit ascent. The chatbot advised them to bring far less food and water than the grueling climb required, recommended simple carbohydrates over fats on the grounds that fats "take too long to digest," and failed to provide adequate warnings about strict turnaround times. The group reached the summit at 7:00 p.m., began their descent in darkness, one hiker fell and injured his knee, and they spent a freezing night in Mud Creek Canyon. Siskiyou County Sheriff Jeremiah LaRue was blunt: people are "over-relying on AI" and need to "go to talk to actual people who know how to hike."

Mount Shasta was not an isolated incident. Around the same time, 16-year-old Bryce Gowryluk of Vancouver used Claude to plan a climb on Crown Mountain, just north of Vancouver in British Columbia. The chatbot suggested Crown Gully, a class-4 scrambling route. The teenager ended up stranded at the base of the Widowmaker's final headwall — a 5.9 climbing route with terrain far more demanding than he anticipated. A helicopter rescue was required. While Gowryluk later said he also consulted Google Maps and the hiking app AllTrails, the AI's role was central to the rescue narratives and raised questions about how far automated confidence can take someone.

The phenomenon of 'cognitive surrender'

Chris Flathmann, a human-centered computing professor at Clemson University who has studied AI and search-and-rescue dynamics, coined a term for this behavior: "cognitive surrender." When humans work deeply with AI systems, they can end up surrendering their own critical thinking. "When AI pretends to be an expert on an unfamiliar topic, you can't call BS on that, because you don't have the expertise," he explained. And once you stop thinking critically, it becomes very difficult to restart — especially in a stressful backcountry situation where rapid decision-making is essential.

This phenomenon is not new to recreational technology history. Roughly a decade ago, social media platforms like Instagram were already being cited as factors in backcountry accidents — tourists following photos of dangerous locations without considering real-world conditions. Before that, Google Maps had been blamed for navigation directions that led hikers onto wrong trails. Generative AI is simply the latest tool in a continuous line of technologies that offer convenience in exchange for agency.

Alarming numbers

Available data suggests that AI use for outdoor recreation planning is growing exponentially. Chip Jenkins, superintendent of Grand Teton National Park in Wyoming, reported that the percentage of visitors using AI for travel information jumped from zero to 26 percent in a single year. "I'm guessing that curve is not going to flatten," Jenkins told High Country News. In Grand Teton, there has been a marked increase in the number of people who report using AI for trip planning.

Despite this, rescue statistics have not yet shown a generalized uptick. At Mount Shasta, the rescue office overseen by Nick Meyers averages a dozen rescues per year, and 2026 appears to be an average year. In Wyoming, rescue calls fell from 466 in 2025 to 312 so far in 2026, compared to 377 in the same period of 2024. But experts warn that the correlation between AI use and accidents may take months or years to materialize in statistical data — which does not diminish the value of the warnings rescue teams are already issuing.

What differentiates generative AI from social media or navigation apps is the persuasive nature of its responses. Chatbots like Gemini, Claude, and ChatGPT are designed to sound confident and authoritative even when providing incorrect information. They "hallucinate" — presenting fabricated details as factual — and tend to tell users exactly what they want to hear. A model might generate detailed routes and supply advice that sounds plausible but is scraped from Reddit threads and outdated YouTube videos, or even be entirely invented.

The consistent recommendation from rescue teams is clear: never rely exclusively on AI for trail or mountaineering planning. Consult official sources (forest service stations, avalanche centers), use multiple navigation methods, and above all, maintain critical thinking. The convenience of AI is not worth someone's life.

Sources: High Country News, PCMag, New York Times

✓ Independent sources cross-checked and verified before publishing