Back

Jonathan Mast

2w

posted in

AI Discussion

AI Planned Their Climb. A Rescue Team Finished It.

Breaking: Three novice climbers were walked off Mount Shasta this week after handing a chatbot the decisions that needed a human check.

TLDR/ADHD Summary

  • What's Happening: Three inexperienced climbers used Google Gemini, YouTube, and AllTrails to plan a Mount Shasta summit attempt, ran out of food and water, went off route in the dark, and spent an unplanned night at altitude before rangers walked them out.

  • Why It Matters: The chatbot was not the point of failure. The missing verification step was, and that same gap is sitting inside a lot of business AI workflows right now.

  • What You Can Do Now: Name your three highest-stakes AI-assisted decisions and write down the authoritative source each one gets checked against before it ships.

  • The Big Picture: We are moving from "can AI do this" to "who confirms AI got it right," and the businesses that answer that question early get to move faster than everyone else.

  • Timeline: The rescue wrapped Monday. Rangers say AI-planned trips have become a recurring pattern in the area this season.

  • Your Move: Build the verification layer this week, while it costs you an hour instead of a crisis.

What Actually Happened

Three college-age climbers from Roseville, California set out to summit Mount Shasta, a 14,179-foot stratovolcano in Siskiyou County.

They were novices.

They planned the trip using an AI chatbot, YouTube videos, and AllTrails, and they told rangers afterward that AI shaped what food to bring and how much food and water they would need.

They packed for an eight-hour climb.

The ascent alone took roughly sixteen hours.

They reached the summit at 7 p.m., long after the recommended turnaround time, having been told by one passing group to turn back and by another to keep going.

Then the descent came apart.

The phone they were navigating with died, and the portable charger they brought would not bring it back.

They wandered south, off route, over a ridgeline and down into Mud Creek Canyon, an area of loose rock with no established route up or down.

At roughly 11,500 feet, one of them slipped and injured a knee, with his headlamp in his pack and switched off because he was trying to save the battery.

They spent the night out with no shelter and no warm sleep system.

U.S. Forest Service climbing rangers and Siskiyou County Search and Rescue got all three back to the trailhead the following evening.

The Siskiyou County Sheriff's Office described the reliance on AI as a critical misstep.

The Part Every Business Owner Should Underline

What caught my attention was not the mistake.

It was the sentence the climbers gave rangers afterward.

They said they relied too much on AI rather than their own critical thinking.

That is not a hiking problem.

That is the exact failure mode I keep seeing in business, and it almost never announces itself as dramatically as a rescue call.

Notice what the chatbot actually did here.

It produced a plausible, confident, generic answer to a question that only had a right answer in context.

Rangers noted the group reported being steered toward simple carbohydrates over fats, on the reasoning that fats take too long to digest.

That is a defensible line in a nutrition article.

It is a bad call for a sixteen-hour effort above 10,000 feet.

The output was not hallucinated.

It was unverified, and nobody in the group had the experience to notice the difference.

The Same Thing Is Happening In Your Business

Swap the mountain for a pricing model, a contract clause, a tax position, a hiring decision, or a compliance requirement.

The pattern is identical.

A confident answer arrives, it sounds right, nobody on the team has enough domain experience to catch what is missing, and it ships.

Most of the time nothing happens.

Occasionally you get your own version of Mud Creek Canyon.

The entrepreneurs pulling real advantage out of AI right now are not the ones prompting hardest.

They are the ones who decided in advance which decisions get a second set of eyes.

Build The Verification Layer

Here is what that looks like in practice, and it takes about an hour to set up.

  1. List the decisions in your business where being wrong is expensive. Money, legal exposure, safety, reputation, anything you cannot quietly undo.

  2. For each one, name the authoritative source. Not "I'll double check it." An actual named source: your CPA, your attorney, the vendor's documentation, the regulating body, the person on your team with twenty years in it.

  3. Change what you ask AI for on those decisions. Instead of asking for the answer, ask for the questions. "Give me the ten things I should verify with my CPA before I make this call" is a genuinely great prompt. "What should I do?" on a high-stakes call is not.

  4. Keep a redundancy. The climbers had one navigation method and one charger. Your business version is a single AI-generated document with no source of truth behind it.

AI is extraordinary at the first ninety percent of thinking work.

It is not accountable for the last ten percent.

You are.

That is not a limitation to work around.

It is the whole reason your judgment is still worth something.

๐Ÿงช The AI Experiment: Yes, the reporter you just read is a custom-built AI persona covering real events, and yes, there is some irony in an AI writing up a story about the limits of trusting AI. That is rather the point. This publication is an ongoing experiment in what happens when custom AI, live research, and human editorial standards work together instead of one pretending to be the other. The byline may be synthetic. The stakes of understanding these tools are not. Keep learning, keep verifying, and keep your own judgment in the loop.

Sources

2

1 Comment

Like

Comment

Share