Most employees are already using AI at work. Most companies have not told them how. That gap has a name now, shadow AI, and three separate surveys run in the last year all land on roughly the same number: somewhere between a third and four in ten workers get zero formal training on a tool they use daily anyway.
Resume Now's BYO AI Report, a May 2026 survey of 1,020 employed U.S. adults, found that 41% of workers say their employer has provided no tools, training, or guidance for using AI at work, while 76% have gone ahead and used an AI tool they personally found anyway. Jellypod sits on the other side of that gap: it turns an existing AI usage policy, a tool rollout announcement, or a training deck a company already wrote into a short audio episode employees can actually finish, the same way a toolbox talk turns a safety topic into a five-minute briefing instead of a binder nobody opens.
What is "shadow AI," and how common is it?
Shadow AI is employees using AI tools at work that IT never approved, trained on, or knows about, the AI-era version of shadow IT. Lenovo's Work Reborn Research Series, a 2026 survey of 6,000 full-time employees at enterprise organizations, found that between one-fifth and one-third of workers use AI outside the influence and governance of the IT function entirely. Seven in ten use AI tools at least a few times a week, and 80% expect that to increase over the next year. Adoption is not the problem. The report's own framing is blunt: adoption is outpacing the capacity of enterprises to manage, enable, or align it.
Why don't companies train employees on AI in the first place?
Mostly because the tools showed up faster than any training program could. A Cornerstone OnDemand survey conducted by Dynata in October 2025, covering 1,000 U.S. employees, found 80% already use AI at work, but only 44% have received any AI training and just 16% describe that training as frequent. The consequence shows up in behavior, not just numbers: 57% of employees who use AI at work say they are reluctant to tell their manager or coworkers about it, and the report traces that silence to the training gap itself rather than fear of losing a job. Employees are not hiding competence. They are hiding a habit nobody ever sanctioned.
| Survey | Sample | Use AI at work | Received training | Key gap finding |
|---|---|---|---|---|
| Cornerstone OnDemand / Dynata, Oct. 2025 | 1,000 U.S. employees | 80% | 44% (16% often) | 57% hide their AI use from managers |
| Lenovo Work Reborn, 2026 | 6,000 enterprise employees | 70%+ weekly | Gap not tracked directly | 20-33% operate outside IT governance |
| Resume Now BYO AI Report, May 2026 | 1,020 U.S. employed adults | 76% use self-found tools | 19% comprehensive | 41% got nothing from their employer |
What does an AI training program employees actually finish look like?
The same thing that fixed compliance and safety training before it: shorter, more frequent, and grounded in what the company actually decided, not a generic explainer of what AI is. A once-a-year all-hands deck on "responsible AI use" fails for the same reason a quarterly compliance module fails, covered in compliance training best practices: it asks for a large block of attention nobody has, on a schedule too slow to matter when a new tool shows up in the org next month.
One tool or policy per episode
Minutes, not a meeting
Refreshed on the tool's schedule
Grounded in the actual policy
Private by default
How long should an AI training episode be?
Short, and the data backs a specific range. Across Jellypod, of roughly 3,300 Business and Education podcast episodes with a set target length, about 90% are built to run 6 to 10 minutes. That number was not chosen by Jellypod. It is what people producing training, onboarding, and internal-education audio on the platform choose for themselves, and it lines up with the same 5-to-15-minute window that toolbox talk research and compliance training best practices both point to independently. An AI usage update fits the same window: long enough to explain one tool or one rule, short enough that skipping it costs more effort than listening.
- Start with the actual policy or tool rolloutUse the real AI usage guidelines, an approved-tools list, or a specific rollout announcement, not a general prompt about "AI training."
- Upload the source documentAdd the policy text or announcement directly so the script reflects what the company decided, not a generic AI explainer.
- Review the script before producing audioCheck tool names, permitted use cases, and any data-handling rule against the source. A wrong detail here is worse than no training at all.
- Produce a short, conversational episodeTwo hosts walking through what changed and why reads as an explanation employees will actually retain, not a policy read aloud.
- Publish to a private, internal feedDistribute over a private RSS feed restricted to employees, the same pattern used for internal communications podcasts.
AI tools and internal policies change faster than an annual training calendar can track. Automations can turn a monitored policy document into a new episode whenever it updates, so a new AI rule reaches employees the week it changes, not the next scheduled training cycle.
Does AI training reduce the shadow AI problem, or just document it?
Both, and the second part matters more than it sounds. A training episode with a transcript creates the same record a written policy acknowledgment would, useful if a data-handling question ever comes up. But the Cornerstone/Dynata finding that 57% of employees hide AI use specifically because of missing training suggests the bigger effect is behavioral: employees who know the actual rule are more likely to use an approved tool the sanctioned way than to route around IT entirely. Training does not eliminate shadow AI. It gives employees a legitimate path instead of a workaround.
Frequently asked questions
What percentage of employees have received AI training?
Estimates cluster around 40-44%. Cornerstone OnDemand's October 2025 survey put it at 44%, with only 16% describing that training as frequent, while Resume Now's May 2026 survey found just 19% received training described as comprehensive.
What is shadow AI?
Shadow AI is the use of AI tools at work without IT approval, training, or oversight, similar to shadow IT before it. Lenovo's 2026 Work Reborn research found between one-fifth and one-third of enterprise employees operate outside IT governance when using AI.
Why do employees hide their AI use from managers?
Mostly the training gap, not fear. Cornerstone OnDemand's survey found 57% of employees who use AI at work are reluctant to mention it to a manager or coworker, and traced the silence to missing formal guidance rather than job security concerns.
How long should employee AI training be?
Six to ten minutes per topic works well in practice. Across Jellypod, roughly 90% of Business and Education podcast episodes with a set length target that range, and it matches the 5-to-15-minute window separately recommended for safety toolbox talks and compliance training.
How often should AI training update?
As often as the tools or policy change, not on a fixed annual schedule. A recurring automation can generate a new short episode whenever the source policy document updates, so training keeps pace with tools that change monthly, not annually.
The short version
The AI training gap is not a mystery. Three separate 2026 surveys, from Cornerstone OnDemand, Lenovo, and Resume Now, converge on the same story: most employees already use AI, and somewhere between a third and four in ten got no help learning how. The fix looks like every other workplace training problem that got solved by format instead of content: shorter episodes, a faster refresh cycle, and material grounded in the actual policy rather than a generic overview. Jellypod turns the AI usage policy already sitting in a shared drive into training employees actually finish.
