Shadow AI culture is what happens when employees stop asking permission and step through the frame.
getty
Something is happening beneath the surface of every industry, and almost no one in a position of authority is paying attention. It started with shadow AI. It didn't stay there.
A parallel economy is forming. Not in crypto. Not on the dark web. In plain sight, but just out of reach of the systems designed to monitor, curate, and control what people do, say, make, and think. It is happening in music, media, journalism, enterprise software, creative work, and social life. It is being built by people who have quietly concluded that the institutions claiming to serve them are actually built to process them. And the speed at which this underground is growing should concern every business leader, investor, and policymaker who assumes the current order is stable.
I call it shadow culture. It started with shadow AI, employees quietly bringing their own tools to work because their companies couldn't figure out how to provide them. But it has expanded far beyond the enterprise. AI is both the catalyst and the accelerant.
To understand what's happening, you have to understand what changed. Two things arrived at the same time.
First, AI tools became powerful enough that a single person with taste and judgment could do work that previously required institutional infrastructure. Record a song. Distribute it globally. Publish a newspaper. Build a software product. Design a brand. The tools are here. They are cheap or free. And they don't require anyone's permission.
Second, the institutions responded to this moment with the worst possible instinct: control.
The platforms that govern digital life have deployed AI systems that not only deliver content but also decide what content exists in your world. TikTok's For You Page, Google's search rankings, Amazon's recommendations, Spotify's playlists: these systems don't tell you what to think. They curate what you're capable of thinking about. They shape imagination upstream of conscious choice.
Governments moved from debating AI to weaponizing the surveillance infrastructure it makes possible. Palantir Technologies secured more than $900 million in federal contracts in 2025. Their ImmigrationOS platform enables real-time tracking of individuals. Scotland Yard is using Palantir’s AI tools to profile its own officers’ behavior patterns. Alex Karp, the company’s CEO, warned publicly that tech firms that don't cooperate with government surveillance risk nationalization. The IoT future that technologists spent a decade promising has arrived without anyone declaring it. Every manufactured object is becoming networked, coordinated, and centralized. Data flows upstream. Control flows downstream. The surveillance architecture is so ambient it feels like the weather.
And enterprises responded to AI adoption by locking everything down, banning tools, blocking URLs, issuing policy memos, and treating every unsanctioned use of a thinking machine as a compliance violation rather than a signal that their workforce had outgrown them.
The combination is explosive. The tools for independence arrived at the exact moment the institutions made dependence intolerable.
So people left.
Here is an uncomfortable comparison that every enterprise executive should sit with: Instagram is easier to use than your company's internal tools.
That sounds like a joke. It is not. According to Microsoft's 2025 Work Trend Index, the average knowledge worker spends 57% of their time communicating, in meetings, email, and chat, and only 43% creating. McKinsey found that they spend 28% of their week managing email alone. Microsoft's data shows they are interrupted every two minutes during core work hours, 275 times per day. Asana's research puts it even more starkly: 60% of a knowledge worker's day is consumed by coordination rather than the skilled, strategic work they were hired to do. Across multiple studies, the losses compound: 103 hours a year in unnecessary meetings, 209 hours on duplicated work, 127 hours regaining focus after interruptions. And the enterprise tools they've been given to manage all of this were designed around administrative compliance, not around the way human beings actually think, create, or solve problems.
Then they go home. They open Claude or ChatGPT. They type a question in natural language. They get an answer in seconds. They draft a document, analyze data, brainstorm a strategy, debug code, write a proposal, and summarize research. The interface is clean. The response is immediate. The experience feels like a conversation with a brilliant colleague. No procurement process. No training module. No IT ticket. No approval chain.
The workflows that most knowledge workers perform every day, such as email, documents, presentations, scheduling, research, and summarization, are not more complex than sharing images and curating a social feed. They just feel more complex because the enterprise tools that wrap them were built for the institution's needs, not for humans. The approval gates, the permissioning layers, the compliance architecture, the audit trails: these exist to serve the organization's anxiety about control, not to help the person sitting at the desk do better work.
When AI tools that actually respect the user's intelligence became available to anyone with a browser, the gap became unbearable. The employee who uses AI at home to do in thirty seconds what takes thirty minutes at work is not a compliance risk. That employee is living proof that the enterprise has failed to build tools worthy of the people it employs.
Ninety-eight percent of organizations report unsanctioned AI use. Three out of four workers bring their own AI tools. The industry calls this "Shadow AI" and frames it as a governance crisis. It is not a governance crisis. It is a product review. The product failed. The users found something better.
Roger Lam, CEO of enterprise AI platform Ordify, sees this firsthand across the companies he works with. "There are two curves shaping the future," he told me. "The exponential acceleration of AI and the linear pace of internal corporate change. Employees are essentially trying to bridge that gap themselves because they feel the treadmill of work accelerating beneath them, while enterprise systems remain frozen in inflexible legacy processes." Lam frames unsanctioned AI use not as a risk but as a signal: a high-fidelity indicator of where an organization's current systems are failing to meet the needs of the people doing the actual work.
Now layer in commuting. According to the U.S. Census Bureau, the average American spends 27 minutes each way getting to work. That's nearly an hour a day, five hours a week, 250 hours a year, spent in transit to a building where they will be interrupted 275 times and spend 60% of their time coordinating rather than creating. Add those commuting hours to the 103 hours of unnecessary meetings, the 209 hours of duplicated work, and the 127 hours recovering from interruptions, and you are looking at a person who loses roughly 700 hours per year, nearly 18 full work weeks, to institutional friction that has nothing to do with the value they were hired to create.
Nobody is measuring this. No annual report includes it. No quarterly earnings call accounts for it. There is an expanding landscape of pay transparency laws requiring salary ranges in job postings and pay equity reporting, but nothing anywhere requires companies to publish how employees actually spend their time by role, location, or function. The transparency movement has been entirely about compensation, not about the quality-of-life toll of how work is actually structured. We have legislated what people get paid. We have not even begun to address what earning that pay costs them. We live in a world where you can buy a biometric tracker on Amazon for nine dollars that will monitor your heart rate, sleep quality, stress levels, and blood oxygen around the clock. We have quantified every dimension of physical health down to the minute. But no company on earth is required to publish how its employees actually spend their time, by role, by location, by function, in a way that would allow a prospective employee, an investor, or a regulator to understand the real toll that work is taking on human quality of life.
Why not? If the point of all this technology is to improve human lives, and not simply to generate an endless feed of AI slop designed to extract attention and money, then where is the transparency? Why isn't there legislation requiring every business, regardless of size, to publish time-use data that shows how work actually happens inside their walls? Not surveillance. Transparency. Give people the ability to compare employers the way they compare products. Let them see which companies celebrate humans doing human things and which ones burn 700 hours a year on institutional friction. Let them vote with their feet.
Right now, the only mechanisms for this kind of transparency are Glassdoor, Blind, Indeed Reviews, Comparably, and their equivalents: shadow message boards driven overwhelmingly by negative sentiment, where HR teams routinely send employees on missions to seed positive reviews and game collaborative filtering scores. The institution's response to reputational transparency is the same as its response to everything else: manipulation rather than improvement. Manage the perception. Don't fix the architecture.
And here is where it gets structural: as AI agents become more capable and more accessible, the power imbalance shifts permanently. It is no longer a question of whether employees can independently do institutional-quality work. They can. The tools are here. The question is whether the institution offers anything worth staying for beyond a paycheck. And for a growing number of people, the answer is no.
If you want to understand why institutions keep getting this wrong, look at the department that manages the relationship between the organization and its human beings.
The term "Human Resources" says it all. It arrived from the United States in the mid-1980s, and even then, observers noted that it treated employees as assets or resources, like machines. But the concept is much older than the label. The first personnel departments appeared in the early 1900s at companies like National Cash Register, created explicitly to manage "the labor problem," a phrase that reveals how the relationship was framed from the start. Frederick Taylor's Scientific Management, the dominant philosophy of that era, treated workers as components to be measured, timed, and optimized. The human being was an input. The factory was the system. Efficiency was the goal.
We have spent 120 years refining the language without changing the architecture.
"Personnel" became "Human Resources." "Human Resources" is now becoming "People Team," "Employee Experience," or "Talent Management." The vocabulary gets warmer with each generation. The underlying structure remains the same: the institution decides what the human being is worth, what tools they can use, how they spend their time, and under what conditions they are permitted to contribute.
COVID exposed this completely. When the pandemic forced remote work, employees didn't just adapt. Many of them thrived. They reclaimed commute hours. They reorganized their days around productivity instead of presenteeism. They reconnected with families, communities, and creative pursuits that office life had crowded out. Digital technology allowed people to time-shift and place-shift, to exercise authority over their own presence and attention. For millions of workers, it felt like a recovery of something essential that the office had quietly taken from them.
The corporate response was revealing. Very few companies used the disruption to reimagine what work could become. Instead, they demanded a return to the office. Not because the data supported it. MIT Sloan found that return-to-office mandates don't improve financial performance but do damage engagement and drive out the highest performers. When Dell mandated a return, nearly 50% of their workforce chose to remain remote rather than accept promotion restrictions. National office vacancy rates remain near 20% despite mandates across every major company in America. The buildings are still empty. The policies didn't work.
But here's the part nobody wants to say out loud: the mandates were never about productivity. They were about control. Companies had made massive investments in real estate and physical infrastructure. Overnight, those investments became worthless as people discovered there were different ways to work. Instead of adapting to the new reality, institutions tried to reverse it. They used policy to override what the workforce had already learned about itself.
This is the same pattern playing out with AI. Employees discovered tools that made them dramatically more capable. Ninety-eight percent of organizations now report unsanctioned AI use. Three out of four workers bring their own AI tools. The institution's response: ban it. Lock it down. Write a policy. Pretend the genie fits back in the bottle.
And the people in charge of enforcing all of this, HR, IT, compliance, and legal, may not be consciously choosing to be instruments of institutional control. But that is what they are. Whether they call themselves the People Team or the Employee Experience Division, they are operationalizing the same philosophy that Frederick Taylor articulated 120 years ago: the institution decides what the human is for. And anyone who deviates is a problem to be managed, not a signal to be read.
But it doesn't have to be this way. If you're a leader reading this and recognizing your own organization in these patterns, here is what the companies getting it right are actually doing.
The humans went somewhere better. That is the part the institutions keep missing. The underground is not a protest. It is a renaissance.
In music, independent artists are building an entire parallel industry. They record without studios, distribute without labels, and monetize through direct fan relationships, community funding, and licensing deals negotiated on their own terms. Artists own their masters. They control their release schedules. They answer to their audience. The major labels still control the playlists. But the infrastructure being built beneath them is designed to make the labels unnecessary.
In the media, the most talented journalists are leaving legacy newsrooms to build direct subscriber relationships on platforms that eliminate the corporate intermediary. When Dave Jorgenson, the Washington Post’s most popular content creator, left to launch his own operation, Local News International, the Post's YouTube views dropped 85% within two months while his new channel outperformed the institution he'd left. Fox News fired Tucker Carlson when he was hosting the highest-rated program in all of cable news, averaging over three million viewers nightly. The network's 8 pm ratings dropped by more than half, Fox lost $800 million in market value in a single day, and Carlson simply moved to another platform and kept his audience. In both cases, the audience didn't vanish. The audience followed the human it trusted and left the institution behind. The Reuters Institute calls this a structural shift. The blunt version is that institutional media is being outperformed by individuals with laptops and conviction.
In creative work, artists are using tools like Nightshade and Glaze to embed invisible alterations in their digital output that poison AI training data. To the human eye, the work looks exactly as intended. To the machine, the data is corrupted. This is not Luddism. These are technically sophisticated creators who understand AI deeply and have chosen to use that understanding to protect their sovereignty.
In social life, people are migrating to Signal, private Discord servers, and platforms designed to be invisible to algorithms. They are building communities where proof of personhood is the price of entry. Where AI-generated content is banned. Where conversation happens without surveillance.
In enterprise after enterprise, employees are quietly building their own AI workflows, solving problems their companies won’t or can’t solve, creating value that shows up in outcomes even when it violates policy.
In every case, the pattern is identical. The institution failed to earn the trust of the people who create its value. So the people left and built something with higher standards.
This is not new. It is one of the oldest patterns in human civilization.
When the Roman Empire made traditional Christian and Jewish burial practices illegal, the early Christians didn't riot. They carved a hundred miles of tunnels beneath Rome and continued doing exactly what they'd always done, just out of sight. They used symbols the authorities couldn't read. They built a complete culture underground: worship, community, art, identity. The empire that drove them there is gone. The tunnels remain.
When the Catholic Church maintained the Index Librorum Prohibitorum for four centuries, dictating what the faithful could read, it eventually recognized that such restrictions did not foster discernment. It abolished the Index in 1966, acknowledging that formation, not control, is how you develop people capable of judgment. The Church got stronger after it stopped trying to govern thought.
When Prohibition banned alcohol in the United States, it didn't stop drinking. It built organized crime, created a vast underground economy, and demonstrated conclusively that you cannot legislate away a behavior that people consider fundamental to their lives. The law was repealed. The speakeasies became the model for an entire entertainment industry.
When the pandemic proved that remote work was viable, companies didn't reimagine the office. They mandated a return to it. The mandates failed. The buildings are still empty. The best employees went somewhere that respected their autonomy.
The pattern is so consistent across civilizations and centuries that it should be treated as a law of human behavior: when you make the surface inhospitable, people build below it. Every time. Without exception. And the underground always outlasts the authority that created it.
The people who leave are never the ones you can afford to lose.
The employee who brings unsanctioned AI to work is your most capable problem-solver. The journalist who starts an independent publication is your most trusted voice. The musician who goes independent is your most ambitious artist. The creator who poisons the training data is the one whose work was worth scraping in the first place.
These are not defectors. They are your culture. And you drove them out because you couldn't figure out how to do anything with AI except cut costs, surveil behavior, and extract value from what you still call, in a term that reveals everything, "human resources."
The business impact will be devastating. Leaders will track headcount, quarterly revenue, and compliance metrics while the actual creative and intellectual capital of their organizations walks out the door, starts something better, and never looks back. Lam puts it in concrete terms: "The most significant business cost is fragmented and lost institutional knowledge. When custom workflows and knowledge generated by those workflows are siloed within individual environments, that intelligence leaves with the employee."
Every unsanctioned AI workflow an employee builds is institutional knowledge the company will never own. The platforms will lose creators. The newsrooms will lose journalists. The labels will lose artists. The enterprises will lose their best engineers. And in every case, the institution will tell itself a story about market conditions or generational attitudes, rather than confronting the obvious: they made the surface inhospitable, and the humans went somewhere they could breathe.
I build AI for a living. I believe in this technology. I also believe that the most important things about being human, authorship, originality, judgment, taste, and moral agency do not survive in frictionless environments. They require resistance. They require spaces where not everything is optimized, monitored, and fed back into a model.
The shadow culture forming right now is friction made visible. It is proof that when you compress humans hard enough, they don't flatten. They create. They build communities with higher standards of trust than the institutions they left. They make work with more integrity than the platforms that tried to own it.
This is the argument I make in my forthcoming book, Human After Friction: the things that slow systems down are often the things that keep humans in control. Friction protects originality. Friction preserves authorship. Friction is where human agency lives.
I explored this pattern in deeper historical and philosophical terms in a recent piece on my Substack, tracing the line from the catacombs beneath Rome to the digital underground forming today. The parallel is not metaphorical. It is structural. And it is two thousand years old.
The institutions that understand this moment will stop trying to drag people back to the surface. They will ask what they failed to build that drove people underground in the first place. They will provide AI tools that amplify human capabilities rather than replace them. They will treat their workforce as authors, not resources. They will recognize that culture is the only asset that compounds, and that you cannot generate culture through surveillance, restriction, or control.
The institutions that don't understand this will continue tightening. They will deploy more monitoring. Write more policies. Rename "Human Resources" to something warmer and change nothing underneath. And they will watch, confused, as shadow AI becomes shadow culture, and their most valuable people vanish into an underground that is already more vibrant, more creative, and more honest than anything happening on the surface.
The tunnels are already longer than they think.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。