AI culture drift: when did that become normal?

First, a question

Four questions your people are answering for you.

Deloitte's researchers found the same four questions turning up inside companies this year, unanswered. People still need an answer to work, so they make one up, and the answer they make up becomes the rule. Tick the ones your company has said something about out loud.

Is it cheating if I use AI to do my work?Somebody senior has said this out loud, in a way people could repeat.
What counts as hard work now that AI does the heavy lifting?There is a shared answer to what a good week looks like.
Who is to blame if AI gets it wrong?One name, agreed in advance, rather than worked out afterwards.
If I do not use AI, will I lose my job?Answered honestly, including the part you are unsure about.

Nothing ticked yet. Every one of these is being answered inside your company today, by people guessing, and the guesses are becoming the culture.

Most leadership teams tick one, and it is usually the third. The other three get treated as too vague to answer, which is precisely why the answers people invent are so consistent and so gloomy.

How it turns up

How a rule gets made without anybody making it.

This is the sequence, and it takes about four months. Every step is somebody behaving sensibly with the information they have.

Somebody tries it One person uses AI for a report and it is better and faster.

Nobody says anything No praise, no objection. Which reads as permission, and also as a reason to keep quiet.

A norm forms sideways Two colleagues copy the method. A third decides it is unfair and stops mentioning what they do. this is the step nobody sees

It is now how things work Four months later there is a rule. Nobody wrote it, nobody agreed it, and changing it now costs a conversation.

The third step is the expensive one and it is invisible from a leadership meeting. Somewhere in there a person decided that being open about how they work is a risk, and that decision does not appear in any system you have.

AI culture drift is a company whose unwritten rules changed while nobody was deciding anything. Eighty in a hundred leaders, managers and workers are concerned that colleagues are using AI to appear more productive than they are.1

Quickly, then

Deloitte surveyed more than 9,000 business and HR leaders in 89 countries this year, with separate worker, manager and executive surveys alongside it.1
80% are concerned their co-workers are using AI to appear more productive than they really are.1
42% of workers say their organisation rarely evaluates the impact of AI on people.1
65% believe their culture needs to change significantly because of AI, and 34% already see culture as a direct blocker to their AI plans.1
Just over half call this important or very important. Five per cent say they are making great progress.1

Probably one of these came up recently

Somebody asked, half joking, whether using AI counts as cheating
Two teams have completely different ideas about what is allowed
People have stopped saying how they made something
The values on the wall no longer describe the week you just had

What you take away

The four questions that are being answered without you
How a rule forms in four months, step by step
The research, with the sample and what it cannot prove
What to say in the meeting where you answer them

In plain terms

Deloitte has a good name for it: cultural debt.

Their definition is the negative consequences an organisation builds up by neglecting its culture, in the same way financial debt builds up interest.1 It is a good comparison because of the second half. You do not pay this back at the moment you borrow it.

What gets borrowed here is a set of answers. Every week somebody in your company faces a question about AI that nobody has settled, and every week they settle it themselves, in a way that suits the pressure they are under. The interest is that a year later those private settlements have become how the place works, and undoing one of them now requires a conversation that would have been three sentences at the time.

Culture is the last twenty decisions nobody discussed. Paul Musters

Deloitte's own framing of the mechanism is precise: when organisations do not answer these kinds of questions, workers are left to navigate new ethical and value-based decisions in ways that accumulate cultural debt.1 Note who is doing the work in that sentence. It is not the leadership team.

And this sits at a specific place on the AI Culture Levels, which is why two companies with the same tools end up with different cultures.

01Campfire60%

02Wild West25%Rules made per person

03Blueprint10%

04Engine4%

05Ecosystem1%

Share of companies per level. Level 2, Wild West, is where everybody uses AI and nobody has said what is allowed, so each person writes their own rule and never checks it against anybody else's. Roughly a quarter of organisations sit here. The shares per level come from emaho's own work with clients.

At Level 2 everybody uses AI and nobody has said what is allowed, so each person writes their own rule and never checks it against anybody else's. That is where the drift happens, and it is where roughly a quarter of organisations are.

What the studies found

Everybody sees the problem. Almost nobody is working on it.

Deloitte's 2026 Global Human Capital Trends research ran with Oxford Economics across more than 9,000 business and HR leaders in 89 countries, with separate worker, manager and executive surveys to catch the gap between what leaders believe and what people experience.1 Four numbers from it are worth putting next to each other.

Concerned that colleagues use AI to appear more productive than they are

80%

Believe their culture needs to change significantly because of AI

65%

Say their organisation rarely evaluates the impact of AI on people

42%

Say they are making great progress on it

5%

Four answers from the same survey of more than 9,000 leaders across 89 countries, with separate worker and manager surveys alongside.1

Read the first and the last one together and you have the whole page. Four in five people suspect their colleagues of quietly gaming this, and one in twenty companies is doing anything serious about it. That gap is where a culture goes while everybody is busy.

34%

of organisations already recognise their culture as a direct blocker to what they want to do with AI

1

20%

of American workers felt strongly connected to their company's culture, in a Gallup poll Deloitte cites

1

1st

time since 2018 that trust in employers declined, according to the Edelman Trust Barometer, also cited by Deloitte

1

Those last two come to us through Deloitte rather than from the original reports, so treat them as context rather than as evidence. What they do usefully is set the ground this lands on. The AI questions are arriving in companies where connection to culture was already thin.

The thing that surprised me most

Deloitte notes that trust is eroding in both directions: leaders are also losing trust in their workers.1 That is the part that changes what you should do about it. If this were only a suspicion running upwards you could fix it with reassurance. It runs both ways, which means the fix has to be something both sides can watch happen.

One caution about all of it. This is a broad global survey of leaders and workers, self-reported, and Deloitte sells consulting on exactly the thing it is measuring. The numbers are consistent with what I see in companies of twenty to five hundred people, and they are not proof of anything about yours.

Why it happens

Nothing was decided, so nothing was announced.

The questions sound too vague to answer

“Is it cheating to use AI” has no policy-shaped answer, so it never gets onto an agenda. It gets treated as philosophy. Meanwhile everybody who writes anything faces it on a Tuesday and needs a working answer by lunchtime, and the answer they reach is the one that protects them.

Silence is the fastest signal there is

Nobody notices a norm forming, because the way a norm forms is that nothing is said. A person does something new, no reaction arrives, and both the doing and the not-mentioning get reinforced. That is why the drift accelerates in exactly the period when leadership feels calmest about it.

You cannot see a rule being made when the way it gets made is that nobody says anything. Paul Musters

Nobody is measuring the human side

Forty-two in a hundred workers say their organisation rarely evaluates the impact of AI on people.1 Licences get counted, usage gets counted, hours saved get estimated. What people now believe about each other gets counted nowhere, so the one thing that is actually changing is the one thing with no number attached to it.

Drift has a level as well

Six questions and three minutes, with no name on it. Your level, what it is costing you, and what one step up changes.

Do the Culture Level scan

Do this first

Ask five people what the rule is. Do not tell them there is one.

Pick five people from different teams and ask each of them the same thing, separately: what is our rule about using AI for work that goes to a customer. Write down their exact words. That is the whole exercise and it takes half an hour.

1

Five different answers

The common result, and the one you want to find now rather than in a complaint. Each person has built a working rule from silence, and every one of those rules is defensible on its own terms.

2

Five answers that agree and nobody can say where it came from

Better and stranger. A norm has formed and it is holding, which means it is real. Write it down and decide whether it is the one you would have chosen, because right now it is running your company by accident.

3

Somebody asks why you want to know

Pay attention to that one. It means the question feels risky to answer, which tells you more about where you are than any of the five answers will.

Do it yourself rather than sending it round as a survey. The wording people use when they are guessing is the finding, and a survey turns that into a tick box that tells you nothing.

One warning from having done this badly. Do not correct anybody in the moment. The second you say what the rule should be, you have stopped collecting and started announcing, and you will not get an honest answer from the remaining four.

In the five levels

Five conversations tell you it is happening. They do not tell you where.

Drift concentrates. It runs fastest where people work fastest and are watched least, and that is rarely the team a leadership meeting worries about.

An Operating Profile in use. Personality type and AI level in one profile, with the agents that fit it.

We measure it person by person: how somebody thinks and works, and where they are with AI. What comes back is a distribution rather than an average, and drift shows up as distance between people on the same team. Two colleagues four levels apart, doing the same job, have already stopped meaning the same thing by a phrase like doing it properly.

It also gives everybody the same words. Most of this argument is people describing different practices with the same vocabulary, and a shared way of naming where somebody is takes the moral charge out of a difference that was never about character.

About emaho

emaho measures one Operating Profile per person: personality type and AI level in a single profile. On that we build a personal set of AI agents that fit how that person works, inside the tools they already use. Fifteen minutes to complete, first profile free, built for companies between 20 and 500 people.

See your own profile, free Find your AI Culture Level

Fifteen minutes per person. No credit card, no strings.

The answer

Answer the four questions in public.

Not in a policy document. In a meeting where people can see you say it and can ask a follow-up, because the thing that has been doing the damage is the silence rather than the absence of a rule.

On the first one, say plainly that using AI is expected and that hiding it is the problem. That single sentence retires the cheating question, and it retires it in the only direction that survives contact with reality, since half your company is already doing it. Deloitte's respondents named open dialogue between workers and leaders as the most important action for driving culture, and clear regular updates on how AI affects work second.1 Both of those are things you do out loud.

On the second, replace the definition of a good week. If effort was the measure, say what has taken its place: judgement, the decisions somebody made about what to keep, the thing they caught. People will keep performing effort until you tell them what you are looking at instead.

On the third, name a person per piece of work rather than a principle. And on the fourth, answer honestly, including the part you do not know, with a date when you will know more. An honest uncertain answer ends the guessing. A confident vague one extends it.

Then do the boring part

Repeat all four in six weeks, and again in three months, with an example each time of somebody who did it the way you meant. One announcement is an event that people wait out. Three is a norm, and a norm you chose beats one that formed while you were busy.

What did the five of them say?

If you run the five-person question, send me the five answers. The spread between them is usually more interesting than any of them on their own, and I will tell you what I would do about it. An answer, not a follow-up sequence.

Message me on WhatsApp

What waiting costs

By the time it has a name, it is already normal.

The cost you can see is the one that arrives as a surprise. A piece of work goes out that nobody would have approved, produced under a rule that somebody invented eight months ago because nobody gave them one. The investigation afterwards finds no rule broken, which is the awkward part.

Underneath sits what four in five people already suspect about each other.1 Suspicion between colleagues is expensive in a way that never shows up as a line item: work gets checked twice, credit gets guarded, and the person who found the clever method stops sharing it because sharing it now looks like showing off or like admitting something.

And there is the interest on the debt. Every one of these questions gets more expensive to answer the longer it goes unanswered, because by then you are not setting a rule, you are overturning one that people have been living by. Three sentences in month one becomes a difficult meeting in month eighteen, and the difficult meeting is the cheap version of what comes after that.

What people search for

What people ask about culture drift

What is AI culture drift?

It is a company whose unwritten rules about AI changed without anybody deciding they should. Nothing was announced. People faced questions nobody had answered, each of them settled on a private answer, and after a few months those private answers are how the place works. Deloitte calls the accumulated result cultural debt.

What is cultural debt?

Deloitte's 2026 Global Human Capital Trends research defines cultural debt as the negative consequences an organisation accumulates by neglecting its culture, similar to the way financial debt accrues interest. The comparison is useful because of the delay: you do not pay it at the moment you borrow it, and every unanswered question gets more expensive to answer later.

Do employees think their colleagues use AI to look more productive?

Widely. In Deloitte's 2026 survey of more than 9,000 leaders across 89 countries, with separate worker and manager surveys alongside it, 80% of leaders, managers and workers said they are concerned that co-workers and teams are using AI to appear more productive than they really are. Four in five, about each other.

How do you see where inside your company the norms have drifted apart?

Drift shows up as distance between people on the same team doing the same job. emaho measures one Operating Profile per person, personality type and AI level in a single profile, so you get a distribution rather than an average. Two colleagues several levels apart have already stopped meaning the same thing by a phrase like doing it properly, and neither of them knows it.

Is it cheating to use AI at work?

That is one of the four questions Deloitte found workers answering for themselves because nobody had answered it for them. There is no universal answer, and there is a useful one: say out loud that using AI is expected and that hiding it is the problem. That version survives contact with reality, since much of the company is already doing it.

How does an unwritten rule about AI actually form?

In about four months, in four steps. Somebody uses AI for a piece of work and it comes out better. Nobody reacts, which reads as permission and as a reason to stay quiet. Two colleagues copy the method while a third decides it is unfair and stops mentioning what they do. By then there is a rule that nobody wrote and nobody agreed.

How many organisations are working on the culture side of AI?

Very few, relative to how many see the problem. In the 2026 Deloitte research, 65% believe their culture needs to change significantly because of AI and 34% already recognise culture as a direct inhibitor to their AI goals. Just over half call the impact important or very important, and 5% say they are making great progress.

Do companies measure the effect of AI on their people?

Mostly not. 42% of workers in Deloitte's 2026 survey said their organisation rarely evaluates the impact of AI on people. Licences, usage and estimated hours saved all get counted. What people have come to believe about each other has no number attached to it, which is why the thing that is actually changing is invisible on a dashboard.

What does emaho do about a culture that shifted without anybody deciding?

emaho gives everybody the same words for where they are, which takes the moral charge out of a difference that was never about character. Everyone gets an Operating Profile, personality type and AI level in one profile, and on that we build a personal set of AI agents that fit how they work, so the standard lives in the tool rather than in a document nobody opens. The first profile is free, fifteen minutes per person, built for companies between 20 and 500 people.

How do you find out what your company's real AI rule is?

Ask five people from different teams the same question separately: what is our rule about using AI for work that goes to a customer. Write down their exact words. Five different answers is the common result. Five matching answers that nobody can source is more interesting, because a norm has formed and it is running your company by accident.

Why does a policy document not fix AI culture drift?

Because the damage was done by silence rather than by the absence of a document. A policy that nobody sees anybody act on changes nothing. In Deloitte's research the two actions respondents named as most important for driving culture were creating open dialogue between workers and leaders, and giving clear regular updates on how AI is affecting work and jobs. Both are things you do in public.

Is trust between leaders and employees getting worse?

Deloitte reports it eroding in both directions in 2026: workers trusting employers less, and leaders also losing trust in their workers. It cites Edelman's Trust Barometer finding that trust in employers declined in 2025 for the first time since 2018, and a Gallup poll finding only 20% of US workers feel strongly connected to their company's culture. That two-way direction matters, because a one-way problem could be fixed with reassurance and a two-way one cannot.

What connects to it

Formed in the same silence

Each one is something people decided for themselves because nobody decided it for them.

07

Shadow AI

What people use when nobody has told them what they may use, and why the good ones keep quiet about it.

Read this one 09

The human cost

The same silence, read from the other side, by somebody working out whether they are still needed.

Read this one 08

Workslop

What the suspicion is actually about, with the hours and the cost of it measured.

Read this one

All 25 AI challenges, and what to do about each

Twenty-five things that break inside a company once people start using AI, each with the research behind it and the level where it starts to bite. This one starts to bite at Wild West, and 8 of the others start there too.

Open the full set

Your next step

See where your people already are, before you write any rules.

Rules written for a company you no longer have will be ignored politely. Fifteen minutes per person shows you the actual spread, and the rules almost write themselves after that.

Look at the spread inside one teamTwo people with the same job and two different profiles. That distance is the drift, and it is measurable.
Write the rules for the company you haveRules written for last year's company get read politely and then ignored.

Start with your own profile, free Ask me anything first Plan a 30-minute call

First profile free · no credit card · built for companies of 20 to 500 · you decide what your team gets to see

Not ready to put your team in anything yet? Start with the level of the company instead. The Culture Level scan is six questions, three minutes, and asks nothing of you.

Paul MustersFounder of emaho

Paul Musters

Fifteen years of leadership and team development in Dutch scale-ups. That practice now sits in software: one Operating Profile per person, with agents that actually fit. He writes these pages from what he runs into with clients, not from a research summary.

LinkedIn · paul@emaho.world · WhatsApp

Sources and numbers used on this page

Deloitte, 2026 Global Human Capital Trends, chapter 4, Dealing with AI's cultural debt, by Jason Flynn, Yves Van Durme, Stephen Harrington and Ashley Reichheld, published 4 March 2026. Research conducted with Oxford Economics among more than 9,000 business and human resources leaders across many industries and sectors in 89 countries, supplemented with separate worker, manager and executive surveys and more than 50 interviews with executives and subject matter experts. Source of: the term cultural debt, defined as the negative consequences an organisation accumulates by neglecting its culture, similar to how financial debt accrues interest; 80% of leaders, managers and workers being concerned that co-workers and teams use AI to appear more productive than they are; 42% of workers saying their organisation rarely evaluates the impact of AI on people; 65% believing their culture needs to change significantly given the impacts of AI; 34% recognising culture as a direct inhibitor to their AI transformation goals; just over half considering the impact of AI on culture important or very important while only 5% are making great progress; the four unanswered questions workers face, which Deloitte lists as whether it is cheating to use AI, what hard work means when AI does the heavy lifting, who is to blame if AI is wrong, and whether not using AI costs you your job; the finding that trust is eroding in both directions, with leaders losing trust in workers as well; and open dialogue between workers and leaders plus clear regular updates on how AI is affecting work being named the two most important actions to drive culture. Deloitte cites within the same chapter a 2025 Gallup poll finding that only 20% of US workers feel strongly connected to their company's culture, and Edelman's Trust Barometer finding that trust in employers declined in 2025 for the first time since 2018; both are used here as Deloitte reports them and the original reports were not read directly. Deloitte sells consulting services on organisational culture.

Numbers are quoted as published. The four-step sequence in the recognition block and the quotations attributed to Paul are drawn from client situations rather than transcripts. The share of companies per level comes from emaho's own work with clients.