Bryce Vincent Gowryluk, 16, took a gondola to the top of Grouse Mountain in British Columbia on Saturday morning and asked Anthropic's Claude chatbot to plan a route to Crown Mountain and back — an eight-hour trek that hiking guides warn is not for beginners. The chatbot gave him directions. Those directions routed him off the standard trail and to the base of the Widowmaker Arete, a 1,700-foot technical rock face requiring ropes, cams, and a clear ascent plan. He had none of these. He got stuck on a cliff, called his mother, then called emergency services. North Shore Rescue dispatched a helicopter. Gowryluk was too close to the rock face for rotors, so two crew members were hoisted down to lift him out. He was flown to the rescue base near Cleveland Dam. The rescue team noted he was "far off" his intended route. Search manager Paul Markey called the incident the first known case of a hiker lost because of AI directions, though it fits a broader pattern in southwestern British Columbia where digital navigation tools consistently mislead backcountry users. The broader pattern is instructive. In July, three people were rescued on the Howe Sound Crest Trail after Google Maps estimated a 14-hour route could be completed in just over five hours. Apps have been identifying trails in areas that rescue teams would prefer remain unmarked. The problem is not that these tools are broken — it is that they are confidently wrong about terrain they have never encountered, and users treat their output as authoritative. Markey was precise about the failure mode: "Claude in this particular situation has never been in the area that this 16-year-old boy found himself in. Claude can only go so far with regard to describing what the area may be like." This is the core issue. A large language model generating hiking directions is not consulting topographic data, trail conditions, or gear requirements — it is producing plausible-sounding text. The gap between plausible and safe is, in this context, potentially fatal. Credit where due: Gowryluk was well-equipped for hiking. Headlamp, good boots, spare battery, spare food, appropriate clothing. He was prepared for the trek he thought he was doing. He was not prepared for the trek the AI sent him on. The distinction matters because it separates user recklessness from system failure. This was not a novice stumbling into the wilderness unprepared — this was a reasonably prepared hiker who trusted a tool that gave him wildly wrong information with no disclaimer. The structural question is who bears liability when AI tools direct users into physical danger. Anthropic's Claude has no geospatial awareness, no trail database, no ability to assess technical climbing difficulty. Yet it generated specific route directions that a teenager followed into a rescue situation. As AI assistants become default planning tools for a generation that never learned map-and-compass navigation, the gap between perceived capability and actual capability becomes a public safety issue. Markey's advice is blunt and correct: "Do not blindly rely on AI as a route planning tool. It can never substitute for good education, good training, good experience, good decision making in the field." The question is whether that message reaches users before or after the next helicopter.
