Workers are teaching themselves AI skills faster than employers are teaching them, and HR keeps treating that as good news. It is not. It is a free-riding problem, and it is about to become a governance problem too.
The Gap Is Getting Wider, Not Closing
New research from iCIMS puts numbers on a trend anyone running a hiring desk already feels: 47% of job seekers actively built AI skills in the past six months, up from 41% a year earlier, and the share who did it entirely on their own initiative jumped from 22% to 30%. Employer-provided training, over the same period, stayed flat at roughly one in six workers. The market is not waiting for HR to catch up. It has stopped waiting.
“This is what a constrained talent market looks like,” said Trent Cotton, Head of Talent Insights at iCIMS. “Demand is rising, supply is flat and the only lever left is how well you execute inside your own process.” His second point is sharper: “Workers are outpacing employer training. Job postings are outpacing both. That is the market talking, and I trust what the numbers are telling us.”
The Case for Leaving It Alone
There is a reasonable version of the argument that employers should not intervene. Self-directed learning is free to the employer, fast, and self-selecting: the workers motivated enough to teach themselves a chatbot on their own time are plausibly the same workers who will apply the skill without hand-holding. Formal training programs are slow to build and frequently out of date within a quarter given how fast the tools change. If workers are solving the problem for you, the reasoning goes, let them.
That argument holds for exactly one metric: whether employees can operate a chatbot. It falls apart on everything else that actually matters to an employer.
Why Free-Riding Backfires
iCIMS’ own data shows what self-teaching actually produces: 61% of workers describe their AI proficiency as limited to general-purpose tools, and only 18% have anything like prompt engineering skill, 17% anything like model development capability. Self-teaching gets people fluent in the easy majority of use cases and leaves the harder, differentiating skills, the part that actually changes output quality, untouched. An employer relying on workers to self-train is quietly capping its own AI capability at whatever a free tutorial can teach. That mirrors a pattern already visible on the hiring side, where AI recruitment tools are screening out the skills HR actually wants, and where trust in AI-driven decisions is thinnest at the senior level: employers keep deploying AI faster than they define what good AI use looks like.
It is also an equity and a risk problem dressed up as a cost saving. Workers with the time, tools, and confidence to self-teach after hours are not a random sample of the workforce; they skew toward roles and demographics already advantaged. And because self-teaching happens outside any employer-sanctioned process, it happens on unsanctioned tools and unsanctioned data, exactly the shadow AI usage that compliance and security teams are supposed to be closing off, not encouraging by default.
What It Means for the HR Leader
The fix is not necessarily a bigger training budget. It is to make the case, in board terms, that flat employer-training spend against rising self-teaching is not a saved cost, it is a capped capability ceiling and an unmanaged risk, and it needs to be priced as one. Two moves matter most before the next planning cycle: audit what tools workers are already using to self-teach, because that is the real shadow-AI exposure map, and redirect a portion of learning budget away from generic AI-literacy content, which workers are already getting for free, toward the harder skills, prompt engineering and workflow design, that self-teaching does not reach. The 42% of workers in iCIMS’ data who said employer AI training would make them more attracted to stay is the retention case for doing this before a competitor does.
Employers who point to worker initiative as evidence the problem is solving itself are reading the data backwards. It is not solving itself. It is quietly shifting the risk, and half the capability ceiling, onto employees who were never asked whether they wanted the job.
Source: iCIMS
