A group of three hikers in their twenties summited Mount Shasta in Northern California at 7:30 in the evening, long past the point where a safe descent was realistic. They had underestimated the climb, brought insufficient food, and misjudged their own fitness. By morning, they were lost on a dangerous section of the mountain, and one had injured his knee. When they later reflected on what had gone wrong, they pointed, in part, to their reliance on Google’s AI chatbot Gemini for trip planning. This is not an isolated incident. It is a pattern, and it raises questions that go well beyond hiking.
The Specific Ways AI Fails in the Field
The Mount Shasta cases are instructive precisely because they are varied. The three hikers who summited too late represent one failure mode: AI providing planning guidance that sounds complete but lacks the contextual judgment a local expert would apply automatically. A separate visitor at the same location pitched a tent near a posted “No Camping” sign and refused to move, citing AI as her authority. A third person arrived at a trailhead holding an AI-generated map that placed prominent geographic features on the wrong side of a nearby mountain.
These are not edge cases of people ignoring good advice. They are cases of people receiving confident, specific, wrong information and acting on it.
In British Columbia, a 16-year-old was rescued by helicopter from a ledge on the Widowmaker Arete on Crown Mountain. He had used Anthropic’s chatbot Claude to help plan the trip. The story spread widely across Canada and the United States. The hiker later noted that he had also consulted Google Maps and the hiking app AllTrails, which complicates any single-cause explanation, but the underlying dynamic remains: AI-generated guidance was part of a planning process that ended in a rescue.
Nick Meyers, program manager of the Mount Shasta Wilderness Climbing and Avalanche Center, described the hikers who used Gemini as people who “turned their brains off.” That phrase is more precise than it might seem.
From 0 to 26 Percent: The Scale Is Growing Fast
Grand Teton National Park in Wyoming offers a useful data point. Park Superintendent Chip Jenkins reported that the share of visitors who say they used AI for travel information went from zero to 26 percent over the past year. Jenkins added that he does not expect that curve to flatten.
That shift in behavior is happening faster than the institutions managing public lands can adapt. Search and rescue call volumes have not yet shown a clear upward trend tied to AI use. Meyers noted that his office averages around a dozen rescues per year and that 2026 is tracking as an average year. In Wyoming, search and rescue calls have actually declined compared to 2025. The data does not yet confirm a crisis. But the behavioral shift is real, and the lag between a new technology becoming widespread and its consequences becoming measurable is a familiar pattern.
Chris Flathmann, a human-centered computing professor at Clemson University who has studied AI and search and rescue, introduced a concept that clarifies what is actually happening: “cognitive surrender.” When people work closely with AI systems, they can gradually stop thinking critically on their own. In a familiar environment, that might be recoverable. In a backcountry situation, under stress, with limited information and physical risk, it is not.
Flathmann also noted a structural feature of how chatbots communicate: they tend to tell users what they want to hear, and they do so with authority, regardless of whether the underlying sources are reliable. A chatbot might produce detailed, confident trail directions assembled from Reddit threads, outdated YouTube videos, or material that is simply fabricated. The output sounds like expertise. It is not.
What This Reveals About How We Delegate Judgment
Search and rescue professionals have watched this pattern before. A decade ago, Instagram and other social media platforms drove people toward destinations they were not prepared for, sometimes with fatal results. Before that, overreliance on Google Maps and smartphone connectivity caused its own category of problems. AI is the latest chapter in a longer story about what happens when people outsource judgment to tools that were not designed to carry it.
The distinction Flathmann draws between social media and AI is worth holding onto. Platforms like TikTok or Instagram offer inspiration. AI offers instructions, and it offers them with the tone of an authority. That difference in register changes how people receive and act on the information. Inspiration invites interpretation. Instructions invite compliance.
Recreation attorney and search and rescue volunteer Jim Moss, based in Colorado, put it plainly: more people are outdoors than ever before, driven by years of media encouragement, and it remains unclear how AI-generated advice will influence all of them.
The expert consensus on what to do has not changed. Verify information against authoritative sources: public land agency maps and websites, trail association resources, current guidebooks. Better still, speak to a park ranger, a national forest official, or a professional guide. These are people whose knowledge is current, local, and accountable in ways that a language model’s output is not.
In Short
AI chatbots can produce confident, detailed, wrong information about outdoor environments, and people are acting on it. The problem is not that AI is malicious. It is that AI communicates with an authority it has not earned, and users, especially those unfamiliar with a terrain, have no reliable way to detect the errors. The concept of cognitive surrender captures the deeper risk: the more a person delegates planning to an AI system, the less equipped they are to catch its mistakes when it matters most. For decisions that carry physical consequences, verification through human expertise is not optional.
Based on reporting from Wired.