When AI Advice Goes Wrong in Mountains, Storms, and Elections

An AI chatbot can offer a detailed answer in seconds, but that answer may send someone in the wrong direction when the stakes are high. Hikers have used chatbots to plan mountain trips, emergency officials have asked AI to help write weather alerts, and millions of voters are turning to these tools for political information.

The common problem is not that AI always gives bad advice. It is that the advice can sound confident even when it is wrong, incomplete, or based on unreliable information. In a backcountry emergency, during a dangerous storm, or before an election, that confidence can make people stop checking the answer for themselves.

When a chatbot becomes the trip planner

By the time three hikers called for help from a rocky face on Mount Shasta in Northern California, they had made a series of bad decisions. They did not bring enough food, overestimated their fitness, and kept climbing late into the afternoon instead of turning back.

They reached the summit at 7:30 pm and had to descend in the dark. By morning, they were lost on a difficult part of the mountain, and one hiker had injured his knee. The hikers later admitted that their poor planning and lack of preparation came partly from relying on Google’s artificial intelligence chatbot Gemini.

Nick Meyers, the US Forest Service program manager of the Mount Shasta Wilderness Climbing and Avalanche Center, said his team often sees visitors arrive without enough preparation. “We get plenty of people who come unprepared or don’t do any research at all. To these guys’ credit, they did try,” Meyers said. “But I think they kind of turned their brains off.”

Meyers also described other examples. One visitor pitched a tent near a “No Camping” sign and insisted she had a right to do so because AI told her she could. His team also encountered a man at a trailhead carrying an AI-generated map that placed prominent features on the wrong side of a nearby mountain.

In British Columbia, a helicopter crew rescued a 16-year-old hiker from a ledge in the Widowmaker Arete on the back of Crown Mountain. The teen had used Anthropic’s chatbot Claude to help plan the trip, though he later said he also consulted Google Maps and the hiking app AllTrails. Stories focused on Claude spread across Canada and the US.

Grand Teton National Park in Wyoming has also seen a marked rise in visitors who say they use AI for travel information. Park Superintendent Chip Jenkins said, “It went from zero to 26 percent. And I’m guessing that curve is not going to flatten.”

That growing use has not yet produced a clear rise in rescues. Meyers said his office averages a dozen rescues each year, and 2026 is shaping up to be an average year. Search-and-rescue calls in Wyoming have gone down compared with 2025, with 312 calls so far in 2026, compared with 466 in 2025 and 377 in 2024.

Still, the risk is difficult to measure. Jim Moss, a Colorado-based recreation attorney and search-and-rescue volunteer, said, “Over the last two decades, there’s been a phenomenal push in the media to get more people outdoors, and consequently we have more people outdoors and recreating.” He added, “It’s still unclear how questionable advice from a robot will influence all those people.”

The danger of confident mistakes

Chris Flathmann, a human-centered computing professor at Clemson University, describes one risk as “cognitive surrender.” When people work with AI, he said, “When humans work with AI systems, they can get so deep into it they give up thinking critically for themselves.”

That can be hard to catch when the user lacks expertise. Chatbots can “hallucinate,” presenting misinformation as real, and they may offer detailed directions built from Reddit threads, out-of-date YouTube videos, or information that was fabricated. Flathmann said that when AI acts like an expert on an unfamiliar subject, “you can’t call BS on that because you don’t have the expertise.”

Chatbots also tend to tell users what they want to hear, using an authoritative voice even when the sources behind an answer are unreliable. Social media such as TikTok and Instagram can inspire someone to visit a place, but AI can provide instructions and encouragement, whether those instructions are right or wrong.

“AI isn’t trying to trick people into going up mountains,” Flathmann said, “but I think it’s the natural evolution of what happens when you get a tool like this.” He also warned, “And once you stop thinking critically, it can be hard to start again, especially in a stressful backcountry situation.”

The advice from outdoor experts is simple: check AI answers against up-to-date public-land agency maps and websites, trail association websites, or guidebooks. Better still, talk to a real-live guide or a national forest official before heading somewhere unfamiliar.

Emergency alerts face the same test

The concern reaches beyond hiking. After Hurricane Isaias made landfall on the Gulf Coast as a Category 3 hurricane, a local emergency manager in Alabama used an AI chatbot to create a graphic showing where the storm might threaten the coast with a “sting jet” of high winds. The graphic was not accurate.

Laura Myers, director of the University of Alabama’s Center for Advanced Public Safety, said the manager wanted stakeholders to focus on the risk in that county. “It did what he wanted it to do, but it was not meteorologically accurate,” she said.

Most emergency management agencies do not permit employees to use generative AI, yet Myers and other experts say its use is likely already common in state and local disaster departments. Hundreds of companies market AI tools for flood damage assessment, resource coordination, and public communication.

Emergency managers are the officials responsible for disaster response and recovery. At the state and local level, they form the first line of defense after an extreme weather event, and they often say that “trust is our currency.” Jeannette Sutton, a disaster researcher at the University of Albany, said limited staff and limited resources make that responsibility harder. “They’re losing the opportunity to be that voice to help people to be safe,” she said.

AI may help with part of the workload. A large language model can read hundreds of thousands of alerts in minutes, review research about similar situations, and edit a human alert in seconds. Rob Dale, who runs the emergency management department in Ingham County, Michigan, said that if an official had sent only one alert 15 years ago and suddenly faced a crisis, “the AI is probably gonna do a better job than you are.”

But Dale would still want control over sending the alert. EM1, a startup that produces AI software for emergency managers, has contracts with around a hundred government agencies and school districts. None uses it to automate extreme weather alerts yet; for now, they use it to workshop plans and internal preparations.

The difference between a rough alert and a useful one can be seen in a flood warning. “Breeched banks. The water is rising. EVACUATE NOW. McGee Park open for people and pets, RV.” does not identify who sent it, where the flooding is happening, who needs to leave, or where to find updates. A revised version names San Juan County, tells people in the evacuation area to leave for higher ground, warns them not to walk or drive through floodwater, and points them toward an official update source.

Trust is also at stake in public information

Officials have struggled with unclear warnings long before chatbots appeared. In 2016, adjacent counties in California issued “voluntary evacuation” and “mandatory evacuation” orders for the same wildfire. In Australia in 2019, the phrase “Watch and Act” caused “significant confusion” because people did not know what it meant.

After the 2017 Thomas Fire, a heavy rainstorm caused a deadly mudslide in Montecito, California. The mudslide killed 23 people and injured more than 160 others. The county sent alerts to everyone in the “voluntary evacuation zone,” telling residents to stay alert to changing conditions and be prepared to leave immediately.

Sarah Miller, a professor of emergency management at Pierce College in Florida, said, “Official adoption is pretty low, but people are unofficially using it, absolutely.” They are using personal accounts or free versions, she said. That raises questions about who checks an AI-written message before it reaches the public.

Political information carries the same problem. The Nov. 3 vote marks the first federal election in which millions of voters are actively relying on AI tools for political information. Studies show that chatbots often draw on unreliable sources across the open web, which can spread election misinformation and erode public trust.

People trust information that feels authentic and shows care, said Ginny Katz, founder of HazAdapt. “AI-generated content does not feel authentic.” Whether the subject is a mountain route, a flood warning, or a political claim, the safest habit remains the same: treat a chatbot’s answer as something to check, not something to obey.

Stay connected via Google News

Follow us for the latest AI News