More Than a Third of Workers Admit Hoarding Expertise Out of Fear of AI Replacement

Companies pushing to automate workflows with artificial intelligence are hitting an unexpected wall: the workers they need to train the systems are increasingly refusing to share what they know. A recent survey of 4,000 workers found that 35% are hoarding knowledge specifically because they fear being replaced by AI, while 38% are reluctant to train colleagues in areas they consider personal strengths. The findings, drawn from a Harvard Data Science Review analysis, underscore a growing tension between corporate efficiency goals and employee self-preservation.

The stakes are not hypothetical. In May, Meta laid off 8,000 people and cited AI as the reason, just a month after telling thousands of U.S. employees that software would capture their mouse movements, clicks, and keystrokes to train AI agents on real work patterns. Across the tech sector, more than 175,000 people have been laid off in 2026, with AI’s rapidly improving capabilities cited as the primary driver. For many workers, the message is clear: teach the machine, then be replaced by it.

The problem is not just reluctance; it is the nature of expertise itself. Some knowledge is easy to codify—templates, playbooks, and workflows can be converted into training data without the expert’s cooperation. But the most valuable expertise is tacit. An experienced salesperson may abandon a lead after the third call, but the record will not show that a hesitation in the client’s voice reminded her of three deals that later collapsed. A senior engineer may deviate from standard procedure, but the log will not reveal that she recognized a combination of weak signals pointing to a deeper problem. These insights can become training data only if the expert chooses to make them visible by explaining reasoning, comparing cases, and articulating intuitions.

Organizational psychologists have long documented this behavior under the label “knowledge hiding.” Research suggests that roughly half of employees intentionally withhold or disguise what they know from colleagues, and the dynamic intensifies when AI is involved. A three-wave study of 348 knowledge workers found that collaboration with AI raises fears of job insecurity, which in turn drives knowledge hiding. Companies cannot simply coerce their way to a solution, because compliance can be performative. This is what researchers call “evasive hiding”: the employee attends every session and answers every question, but withholds the information that truly matters.

Observation is not a complete answer either. Meta’s keystroke capture program follows a century-old logic first applied by Frederick Taylor, who brought stopwatches onto machine-shop floors to turn workers’ practical know-how into measurable, standardized processes. The machinists of that era understood the stakes and coordinated around a deliberately restricted pace, discouraging one another from revealing how quickly the work could really be done. The same dynamic is playing out today, albeit with digital surveillance replacing the stopwatch.

The analysis concludes that business leaders must rethink their approach to knowledge transfer. Better retrieval systems, powered by large language models, can search unstructured data and answer questions in ordinary language—an enormous improvement over manual searching. But retrieval only helps with knowledge that made it into the system in the first place. When experts fear that sharing their hard-won insights will cost them their jobs, the system will remain hollow, no matter how sophisticated the AI becomes.

Author

Delaney Sawyer

Society Reporter

Delaney Sawyer covers public affairs, politics, business, culture and daily news for Core Memo. The role focuses on verification, context, and clear explanations for readers.

All articles by this author