Over the past several months, a concerning pattern has emerged from trails and mountains across the world: people getting lost, injured, or stranded in hazardous conditions after following advice from artificial intelligence chatbots. What began as isolated, nearly viral incidents — the Canadian teenager airlifted from a cliff, the three young Californians stranded on Mount Shasta — now reveals itself as a systemic phenomenon that worries researchers, search-and-rescue teams, and park authorities.
The case that generated the most attention occurred in British Columbia, Canada, where 16-year-old Bryce Gowryluk was airlifted after becoming stranded on a technical climbing route known as the Widowmaker Arete on Crown Mountain. The teenager had consulted Anthropic's Claude chatbot to plan his trip — requesting a packing list and validation that the route matched his skill level. What went largely unreported is that Gowryluk also used Google Maps and the AllTrails app for navigation. He himself acknowledged the navigational error: "That was my mistake; that was my error," he told The New York Times. But the story went viral with the narrative that the chatbot had "pushed" him up a technical climbing route for which he was unprepared.
Paul Markey, rescue manager at North Shore Rescue, was unequivocal: "The main thing is: do not blindly rely on AI as a route planning tool," warning that chatbots and navigation apps cannot replace proper training, experience, and sound judgment. The teenager carried gear appropriate for hiking but not for the technical climbing terrain he encountered.
The pattern repeated itself in equally dramatic fashion on Mount Shasta in California. Three 20-year-olds from Roseville consulted Google's Gemini to plan their ascent. The AI advised them to carry far less food and water than they needed, classifying the route as easy when it was strenuous. Their planned 8-hour climb became a multi-day ordeal when they reached the summit at 7 p.m., hours past the recommended turnaround time of noon. On the descent, without adequate lighting, they strayed into Mud Creek Canyon, where one fell and injured his knee. They spent the night on the mountain before rescue teams found them the following day.
Sheriff Jeremiah Larue of Siskiyou County was blunt: "This was a critical misstep," he said in a press release. AI tends to give favorable information that makes you feel invincible, and that causes problems when something can be extremely dangerous. "AI can be a good tool, but it shouldn't be used as an expert. Talk to real people."
The escalation of these incidents is alarming. 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% in a single year. "I'm guessing that curve is not going to flatten," he said.
Nick Meyers, director of the Mount Shasta Avalanche Center, confirmed there have been at least three incidents in 2026 where hikers or campers were led astray by AI or relied too heavily on it. One involved a woman who pitched her tent near a "No Camping" sign, insisting she had the right to be there because AI told her so. Another involved a man wandering a trailhead with an AI-generated map showing prominent features on the wrong side of a nearby mountain.
The phenomenon Chris Flathmann, a human-centered computing professor at Clemson University, calls "cognitive surrender" is central to understanding why these incidents are so frequent. When humans work with AI systems, they can become so immersed that they give up critical thinking altogether. When AI pretends to be an expert on an unfamiliar topic — like mountain terrain — the user cannot spot that it is wrong because they lack the expertise to verify. And once you stop thinking critically, it is hard to start again, especially in high-stress backcountry situations.
Academic research supports this observation. A study conducted by researchers at the University of California, Irvine, and published in the journal Nature Machine Intelligence demonstrated that humans tend to over-rely on AI-generated recommendations, even when those recommendations contradict direct visual evidence. Confidence did not decrease when users knew the AI might be wrong — in several experiments, confidence actually increased when the AI provided information with an authoritative tone.
Chatbots also have a peculiar tendency: telling users what they want to hear, and doing so with authority, regardless of whether the sources they consult are reliable. Claude or ChatGPT might answer a query with extensive directions and advice that sound credible but were scraped from Reddit threads and outdated YouTube videos — or might be entirely fabricated.
What makes AI particularly dangerous in the context of outdoor recreation is that it has no internal mechanism to signal uncertainty. A chatbot that classifies a trail as "easy" when it is actually strenuous sounds just as authoritative as one that is correct. The difference is that, while an honest human guide would say "I'm not sure" when uncertain, AI invents an answer.
The OECD (Organization for Economic Cooperation and Development) has issued similar warnings about travelers being misled to non-existent destinations — such as the "Sacred Canyon of Humantay" in Peru — leading to financial loss and physical harm. The problem extends beyond hikers: travelers using chatbots for trip planning face similar risks in urban and rural contexts.
The question is that AI is now at the intersection of two realities. On one side, it is a powerful planning tool: it can help organize gear, predict weather, and discover new trails. On the other, it cannot understand terrain, altitude, real-time weather conditions, or the difference between a scenic walking path, an unmaintained gully, and a Class 4 climbing escarpment.
As AI becomes increasingly embedded in wearables and portable devices — smartwatches with built-in assistants, earbuds with voice AI, augmented-reality glasses — the boundary between planning and executing dissolves. The average user no longer clearly separates when they are "planning" with AI and when they are "following" it. And in the real world, where the difference between a planning error and an execution error can mean life or death, that distinction is the most important one that exists.
Sources: Wired, High Country News, Business Insider
✓ Independent sources cross-checked and verified before publishing
