A survey of 1,000 AI workers at foundation model labs, startups, and infrastructure companies reveals widespread self-censorship, exit intent, and internal distrust across the industry.

Why This Matters Now

We are eighteen months into the most aggressive hype cycle in the history of technology. Trillions in market cap, a labor market reorganizing itself around three letters, and a steady diet of keynote demos promising the imminent obsolescence of, well, you. Here's the thing about hype cycles: they run on belief, and belief requires someone at the center who actually believes it. 

We surveyed 1,000 people who work at foundation model labs, AI startups, and AI infrastructure companies and asked them anonymously whether they buy into what their employers are selling. The answer should worry the CEOs who are cheering on this boom more than any regulator ever will.

Key Stats:

  • 87% of AI employees would leave their current job for a company with a more honest public position
  • 82% have stayed quiet in an internal meeting about a safety, capability, or accuracy concern because it felt professionally risky
  • 77% say a colleague has privately told them they're considering leaving over the company's direction or honesty
  • 72% personally support a memorandum to pause AI capability advances
  • 72% say they'd follow a senior colleague who resigned citing safety concerns
  • 45% are actively job hunting right now, not just open to it
  • 23% believe their own company's claims about what its AI can do today are overstated

The Believers Have Doubts

77% of AI employees say they fully believe their company's claims about what its AI can do today. That's the headline you'd expect a PR team to lead with. Here's the one they'd bury: nearly 1 in 4 people building the product say those claims are overstated, and almost 2% call them substantially overstated. In an industry built entirely on the promise of what's coming next, a quarter of the workforce thinks the company is already overselling what's here now.

The Marketing Machine Runs on Overstatement

It gets worse once you follow the money to marketing. Among employees whose job actually touches marketing, sales, or communications, 79% say they've personally been asked, in the past 12 months, to help produce material they themselves thought oversold the product. Not "the competition oversells." Not "the industry oversells." Their own material, at their own desk.

A Culture of Enforced Silence

Here's the number that should eventually end up in a congressional hearing: 82% of AI employees have stayed quiet in an internal meeting about a safety, capability, or accuracy concern because raising it felt professionally risky. More than half (54%) say it's happened more than once. Most people at this company have simply decided the risk isn't worth it.

Golden Handcuffs and the Fear Economy

Why the silence? We asked. Fear of being blacklisted in the industry (41%) and fear of losing equity or vesting (40%) beat out NDA clauses (30%) as the top reasons people won't speak publicly about disagreements with their company's AI claims. In other words, the real leverage here is financial and reputational, not legal. Nobody needs a non-disparagement clause when your paycheck and your next job both depend on staying quiet.

Waiting for Someone to Go First

Employee trust is more fragile than any annual report suggests. If a well-known figure at their company resigned tomorrow, citing safety concerns, 72% say they'd follow. Not "consider it". Follow. This is a workforce standing near the exit, waiting for someone credible to open the door first.

Support for Hitting the Brakes

Given the choice, 72% of AI employees personally support a memorandum proposing a pause on AI capability advances. Read that twice: the people whose salaries, equity, and career narratives depend on the AI boom continuing at full speed are, by the majority, in favor of slowing it down. That is not the position of people who think everything is fine.

The Exit Is Already Open

This is the labor-market number that matters most: 87% of AI workers would leave their current job for a company with a more honest public position; 45% are actively looking right now, not someday. 77% say colleagues have privately told them in just the last three months that they're considering leaving because of concerns about the company's direction or honesty. 51% have heard this from more than one colleague.

The Whisper Network Is Louder Than Leadership

81% of employees have had conversations with colleagues about company concerns that they have not raised directly with leadership; 44% say this happens regularly. There is an entire second conversation happening inside every AI company right now, running in parallel to the official one, and leadership isn't in the room for it.

What Comes Next

A talent shortage won't define the AI labor market in 2026. A shortage of trust'll define it. Companies are competing for the same scarce researchers and engineers, while internally, a majority of their own workforce is quietly rehearsing their exit and hoping someone credible goes first. The firms that win the next 18 months of hiring won't be the ones with the biggest compute budget − they'll be the ones willing to say something true out loud before their employees say it for them, on their way out the door.

Methodology: The data in this report derive from a survey conducted by Movchan Agency. The survey was launched in mid-September 2026. In total, 1,000 employees at AI companies completed the survey.