When AI Fear Goes Underground

There is a deeper problem with AI than most people are talking about.

The headlines give us no shortage of things to fear. AI is coming for our jobs. Data centers will swallow the power grid ¹². Our children are bonding with machines instead of people. Privacy is finished ³. And truth itself is at risk, because LLMs produce falsehoods as a matter of design, not malfunction. Pick an AI fear, any fear, and there is a research paper or conference keynote feeding it.

But the threat I think leaders should actually worry about isn’t the machine. It’s the fear of the machine. And unlike data centers or superintelligence, this one is already inside our gates.

In mid-2025, the two most prominent figures in the AI industry told us to brace for catastrophe. Sam Altman warned that some job categories would vanish completely, and Dario Amodei predicted an end to half of all entry-level white-collar jobs ⁴. These were not fringe speculations. This was coming from people building the technology and speaking with the authority of people who should know.

Then, this past week, both of them walked it back. Altman now says he was “pretty wrong” about AI’s economic impact and is “delighted” to have missed the mark. Amodei now says human work will expand, rather than contract ⁵. These reversals happen to coincide with both companies preparing for public offerings valued at nearly a trillion dollars, a detail the financial press has not been shy about pointing out.

So, what about the people actually responsible for steering companies? They are expected to set an AI strategy, justify the spending, and explain to employees, customers, and investors where this is all going. One quarter the mandate is to move fast before AI eats your industry. The next quarter the same people are walking it all back. The whipsaw is real and it is being felt at every level of our organizations, from the boardroom to the front lines.

This is the world our organizations are living in, top to bottom. Not a world of clear danger, which people can face, but a world of whiplash, which we cannot. Everyone is being told to panic and then told to calm down by the same voices, on a timeline that seems to track those voices’ interests rather than the truth. No wonder people are uneasy. They are not being irrational. They are being jerked around by oracles who cannot keep their own story straight.

And here is the part that matters most for business leaders. Our people are not vocalizing their fear. Our best employees do not announce their fear in meetings. It goes underground, into slow adoption, quiet workarounds, and into projects that never launch. Our strategy decks and all-hands meetings don’t reach this fear, in part because we are all working from the same shifting ground.

Franklin Roosevelt, in a moment of genuine national fear, told a frightened country that the only thing it had to fear was fear itself ⁶. He was right then, and I think the line holds now. The real obstacle to our AI ambitions is not the technology, Altman, Amodei, or their ilk. It is frightened organizations, leaders included. Leading frightened people while you are uneasy yourself may be one of the hardest challenges an executive ever faces.

The Tiger in the Room

Fear, of course, is as old as humanity. Older, really. It is hard-wired into biological organisms as our primary threat detection system. We can suppress our fears, but we cannot eliminate them. Humans fear anything that has power over us, anything that can control us, and anything that can replace us. AI manages to be all three at once, so of course we fear it. And, it is worth pointing out that fear is a pattern detector, not a truth detector. It fires on the shape of a threat, whether or not the threat is real.

Psychologists describe fear as a hard-wired sequence of protection ⁷. First, we freeze. Muscles tighten, breathing shortens, attention narrows, and we assess. If we register a real problem, we look to escape, because getting away is almost always cheaper than the alternatives. When escape is impossible, other responses take over. We fight, which is a bid to overpower the threat. Or we fawn, which is submission, showing the threat we are no danger to it and petitioning for mercy and cooperation. And when nothing else is available, the body can simply collapse and play dead, hoping the threat loses interest. Rabbits freeze. Dogs fawn. Possums go limp. Humans, being complicated, can do it all.

The executive leading people through the jungle of AI, and it is a jungle, will see every one of these responses in their people. Some will freeze and wait for instructions. Some will quietly avoid, routing around the new tools. Some will fight, openly resisting or undermining. And some will figuratively play dead, going through the motions while contributing nothing. None of it shows up labeled as fear. It shows up as slow adoption, missed deadlines, and a strategy that does not move forward.

To many in our organizations there is a figurative sabretooth tiger lurking around the office. It cannot be locked outside. It will not go away. It fears nothing itself. Our people are afraid and they need sound leadership. Executives who freeze, or fight, or fawn themselves are not helping their people. They are feeding the fear itself. The executive’s job is to be the animal that does not bolt, the one the others orient to, so the herd moves together instead of scattering. And a herd that faces the threat together is a herd that has, in the end, nothing to fear but the fear itself.

Is Fear a Cause of Death?

The wreckage from the past year is hard to ignore. S&P Global reported that the share of companies scrapping most of their AI initiatives jumped from seventeen percent to forty-two percent in a single year ⁸⁹. After one of the largest waves of technology investment in modern corporate history, a large majority of it has returned nothing the business could point to.

When the post-mortems get written, they reach for the usual causes, and they are not wrong about them. Poor data readiness. No clear success metrics. Projects run as isolated IT experiments that never touched a real workflow. Lost executive sponsorship. And, again and again, low adoption and user resistance. Study after study lands on organizational causes, not technical ones, as the dominant reason these efforts fail.

It seems to me that something is missing from those reports. In every case the named cause of failure is only the most visible one, but fear may be lurking behind them all. Low adoption is what fear looks like from the outside. So is a workaround, a stalled pilot, a tool that sat unused until the budget for it was quietly killed. The failure analyses describe the bruises in careful detail but never mention the fall.

Not one of these studies lists fear as a cause. I find that telling rather than reassuring. We are willing to write down that people did not adopt the tool, that they resisted it, that they routed around it, and we stop one inch short of asking why. People resist what they are afraid of. They route around what they do not trust. The behaviors fill the reports; the feeling underneath them never makes the page.

I am not claiming fear is the only reason AI projects fail. Data really is messy, and metrics really do go undefined. But to deploy a technology your people are quietly afraid of, and then to explain the failures without once naming that fear, is to write the post-mortem and leave out the cause of death. Every one of those failed projects ran through a workforce that was afraid. They must have. The fear was always in the room. It simply never made the report.

Steadying the Herd

We have all seen videos of a herd of antelope being stalked by a lioness on the Serengeti. At the edge of the herd is the sentinel, the watcher, the member of the herd who protects all others. It peers steadfastly at the hunched predator in the grass. The sentinel’s eyes seem to speak, “I see you!” “You can’t fool me!” As long as the herd stays together and the sentinels are alert the herd is protected.

When it comes to our corporate AI, the same model is applicable. First assemble the herd, second station the vigilant sentinels.

Assemble the Herd

After sitting with more than a hundred CEOs, I can tell you that nearly every serious approach to AI organizes itself into some version of three groups.

The AI Council decides where the herd moves. A body of senior executives who set strategy, allocate resources, and own the metrics and accountabilities.

The AI Guild knows the ground. The technical experts and validators who set standards, vet tools, and judge which terrain is safe to graze and which is not.

The AI Practice Group is the herd actually feeding. These are the practitioners doing the real work of adoption, the people who turn a licensed tool into an everyday habit, and whose success or failure is the entire reason the other two groups exist.

Here is where almost everyone goes wrong. They assemble the herd from the top down. Stand up the Council first, let it set strategy, convene the Guild to pick tools, and only then, eventually, get around to the practitioners who were supposed to benefit. It feels orderly. It is also backwards. The technology does not deliver value the moment it is purchased or the moment a strategy deck is approved. It delivers value when one person tries to use it on a real task, gets stuck, and works through it. A Council deciding and a Guild validating produce nothing if the herd never actually grazes.

So start where the value is. Start with a Practice Group. It is the rare first move that asks almost nothing of leadership: no fixed strategy, no public position, no bet you cannot reverse. You put one group into motion then you watch and listen. Done well, it is fast, inexpensive, fully reversible, and it shows you what is already true inside your business, the quiet experimentation your people are doing on their own, brought into the open where it can be seen and shared. Everything the Council would later decide and the Guild would later validate is better for being built on what the Practice Group already revealed.

Now, notice which group does the calming. Not the Council, whose decisions most employees never see, and which can frighten more than it soothes, because nothing unnerves people like knowing their future is being decided in a room they are not in. Not the Guild, where experts reassure other experts. The fear lives at the capillary level, in the individual contributor staring at a blank prompt and wondering what it means for them. So that is the only level where it can be answered. The Practice Group calms the herd not by telling it to calm down, but by putting frightened people shoulder to shoulder with peers facing the same task, until the tiger in each person’s imagination shrinks to the size of the actual problem in front of them. Reassurance is not a feature we add to the Practice Group. It is what the group is.

Enlist the Sentinels

A herd that has stopped scattering still needs someone watching the tall grass. In nature the sentinel is the animal that stands while the others graze, scanning for what the herd cannot see while its heads are down. A Practice Group needs the same role, and filling it well is what separates a group that compounds the value of AI from a lunch-group that fizzles by spring.

The sentinel is the facilitator. Not the most enthusiastic member promoted into the job, but someone who has spent years guiding practitioners, experts, and executives through exactly this kind of terrain. Being a strong member of the herd does not make you a sentinel.

A good one does not stand watch empty-handed. They carry proven modules, a library of durable principles about AI judgment that hold up no matter which tool ships next week. They bring method: session designs, pre-reads, and cases, so people arrive with shared ground to work rather than a cold start. They know the tools, and where each one is safe to graze and where it is not. And they keep a cadence, the steady operating rhythm and follow-through that stretched internal members, taxed by their day jobs, rarely have the time to sustain.

But the deepest reason the sentinel works is the one easiest to miss. A sentinel standing inside the herd, grazing alongside everyone else, shares the herd’s blind spots. The most valuable sentinel comes from outside.

An outside sentinel answers to no one in the field. They can referee a disagreement, challenge a sacred assumption, and bridge two groups that have spent years not talking, in ways no internal leader, embedded in the same hierarchy and politics, ever could. People tell an outsider things they will never say to their own peers or their boss, which means the outside sentinel can name a dysfunction the inside voices have learned to step around. And because this sentinel has stood watch over many herds, not just yours, they import what works from terrain your people have never crossed. Your internal group can only reason from the one company it has lived in. The outside sentinel has seen the predators that have not reached you yet.

Then there is continuity. Herds change. People rotate in and out, champions move on, the membership a year from now is not the membership today. An outside sentinel provides the steady presence that holds the group’s method and memory together across all of it, so the group does not have to rediscover itself every time the faces change.

And the sentinel faces both directions. The best ones watch the horizon protecting the herd and also signal back to leadership, surfacing what is landing, what is stuck, and how people honestly feel about the work. That telemetry is something no strategy deck can capture, because it comes from the grass, not the map.

The Simplest Hard Thing a Leader Will Do

So, we come back to where we started. Leading people through fear is one of the hardest things an executive is ever asked to do, and AI has handed every leader a version of it at the same time. The fear is real, it is underground, and you are feeling your own version of it even as you are expected to steady everyone else.

Here is the strange grace of this particular problem. The fear is enormous, but the first step to confronting it is small. You do not have to predict whether Altman or Amodei had it right. You do not have to author a grand strategy, stake a public position, or make a bet you cannot walk back. You do not even have to make the fear go away, which is fortunate, because you cannot. You only need to refuse to bolt, scatter, and leave members of the herd to their own devices. All you need is one practice group in motion.

That is the whole move. Assemble one Practice Group. Give it a sentinel who knows the terrain. Then watch and listen. It is fast, it is inexpensive, it is completely reversible, and it does the one thing no strategy deck can do: it brings the quiet, scattered, fearful experimentation already happening in your company into the open, where people can face the tiger together instead of each alone in the tall grass.

Rarely does a leader get a simple, elegant answer to a daunting, high-stakes problem. This is one of those rare times. The problem is hard, but the answer is dead simple.

That is the work we do at Seampoint. We help you assemble the herd and we stand watch as your sentinel, bringing the modules, the method, and the cadence so your people reach competence and confidence faster than they ever would alone. If you have green shoots already, an informal channel, a brown-bag group, a handful of believers, or not, we will help you turn what you have into something that compounds and protects.

Give us thirty minutes. We will find where your people are already experimenting, show you how we would steady and equip them, and help you take the one small step that makes everything after it possible.

Endnotes

1. 1. Belfer Center for Science and International Affairs. “AI, Data Centers, and the U.S. Electric Grid: A Watershed Moment.” February 9, 2026. ↩
2. 2. MIT Energy Initiative. “Data Center Power Demand.” February 26, 2026. ↩
3. 3. Stanford Institute for Human-Centered Artificial Intelligence. “Privacy in an AI Era: How Do We Protect Our Personal Information?” March 17, 2024. ↩
4. 4. Amodei, Dario. “AI jobs danger: Sleepwalking into a white-collar bloodbath.” Axios, May 28, 2025. ↩
5. 5. HR Executive. “OpenAI, Anthropic CEOs Walk Back AI Job Warnings as IPOs Loom.” May 28, 2026. ↩
6. 6. Franklin D. Roosevelt. “First Inaugural Address of Franklin D. Roosevelt.” Yale Law School, Avalon Project, 1933. ↩
7. 7. APN. “Fight, Flight, Freeze, Fawn, and Flop: Responses to Trauma.” November 14, 2021. ↩
8. 8. S&P Global Market Intelligence. “AI Project Failures Surge to 42% as Companies Struggle to Scale.” March 18, 2025. ↩
9. 9. This Week in Health IT. “AI Project Failures Surge to 42% as Companies Struggle to Scale.” March 18, 2025. ↩