57% of Your Team Is Hiding Their AI Use From You

I saw the 57% figure first in a Harvard Business Review piece, and my instinct was to distrust it until I found where it actually came from. It traces to KPMG and the University of Melbourne Business School's 2025 global study, Trust, attitudes and use of artificial intelligence, which surveyed more than 48,000 people across 47 countries between November 2024 and January 2025. KPMG's own release states it plainly: "over half (57%) of employees say they hide their use of AI and present AI-generated work as their own." Not a vibe, not an inherited paraphrase – a named institution's own number, on a sample large enough to take seriously.

Most of the coverage I've seen treats this as a workplace-culture story – psychological safety, employees afraid to look replaceable, the usual "build a culture where people feel safe admitting X" advice. That's not wrong, exactly – it's also not the version of this stat that should worry you most if you run a firm whose entire product is judgment.

Worth sitting with the scale of the sample for a moment before moving past it. This isn't a survey of a few hundred people at three companies who happened to answer a LinkedIn poll. It's 48,000-plus people, across 47 countries, over a three-month window – large enough that "maybe this doesn't apply to my team" is a harder position to hold than it feels like it should be. A number this consistent, at this scale, is describing something structural about how people relate to AI tools at work right now, not a quirk of one industry or one company's culture.

The number that explains the number

Two more figures from the same research, reported via Forbes' coverage of the KPMG dataset, make the 57% much less mysterious. Employees favor public, unsanctioned AI tools over anything their employer actually provides – about 70% use public tools at work, compared with only 42% who use tools the company issued. And only 40% of employees say their employer has a clearly communicated AI usage policy at all. Put those together and the concealment stops looking like a trust problem and starts looking like an obvious response to a vacuum: people are reaching for whatever AI tool is in front of them, mostly not the sanctioned one, inside organizations that mostly haven't told them clearly what's allowed – and then not mentioning it, because there's no clear frame in which mentioning it helps them.

I want to be careful about what this data does and doesn't support. It doesn't tell us employees hide AI use specifically out of fear of losing their jobs, or any other single motive – that's a plausible read, but it's not what was measured. What was measured is a policy-communication gap sitting right next to a concealment number, and the two are clearly related even without a stated causal mechanism.

Why this is a governance problem, not a culture problem

Here's the reframe I think matters most for a boutique advisory practice or an independent consultancy. Most coverage of this stat is written for large-enterprise HR audiences worried about morale and psychological safety. That's a real concern, and it's not the one that should keep a principal at a small firm up at night. The one that should is this: if you sell judgment – if clients pay you specifically because you stand behind the analysis, the recommendation, the number in the deck – you cannot price, audit, or vouch for work whose actual AI-assistance level you don't know. Concealment isn't just uncomfortable; it's a direct hole in the thing you're actually selling.

I've written before about the ungoverned AI habits already sitting inside a lot of teams – tools nobody vetted, being used on work nobody's tracking. This KPMG data is the missing half of that picture: it's not only that unsanctioned tools are in use, it's that more than half the people using any AI tool, sanctioned or not, aren't telling anyone. If your due-diligence and quality-control processes assume you know where AI touched a deliverable, this stat says that assumption is wrong more often than not.

That's a genuinely different problem from the one I wrote about in whether telling clients you use AI costs you their trust. That piece was about the client-facing disclosure decision – what happens to trust when you tell them. This is upstream of that entirely: it's whether you, running the practice, actually know what your own people did before you ever get to decide what to tell a client. You can't make a good disclosure decision about something you don't have visibility into in the first place. And due diligence built to catch AI-washing on the buy side has the same blind spot in reverse – it's checking whether a target overstated its AI use, while this stat says the more common failure inside most organizations is understating it, quietly, at the individual level.

One note on the number itself, because I'd rather flag it than have you find it and wonder if I missed it: you may also see a 61% figure attached to a similar-sounding claim in some coverage of this same study – "avoided revealing use of AI," rather than KPMG's own worded "hide their use and present as their own." Those may be two different survey questions inside the same research, or a rounding inconsistency between outlets; nothing I found resolves it cleanly. I'm using KPMG's own verbatim 57% here because it's the number the institution itself put in its own release, not because the discrepancy doesn't exist. If you go looking at the underlying report yourself, don't be surprised to see both.

What a policy gap this size actually costs you

The 40% figure is the one I'd act on first, because it's the most fixable. A policy that exists but isn't clearly communicated is functionally the same as no policy at all, from the employee's side of it – and this study suggests that's the more common state than an actual absence of any policy. If you haven't checked recently whether your team could describe, in one sentence, what's approved and what isn't, you likely have the same gap KPMG measured at a global scale, just at whatever headcount you run.

That's a cheap thing to test, and most firms never do. Ask three people on your team, separately, what AI tools they're allowed to use on client work and what they have to disclose when they do. If you get three different answers – or a confident answer that turns out to be someone's guess rather than something you actually wrote down and circulated – you've just found your version of the 40% gap. It costs nothing to check and it's uncomfortable to learn, which is probably why most firms skip it.

None of this means assuming the worst about your own people. It means assuming the stat applies to you until you've actually checked, because a number this large, from a sample this size, isn't describing someone else's team. Fifty-seven percent isn't a niche behavior. It's the median.