Deloitte's 2026 Global Human Capital Trends report, released this month, ran a single question past leaders, managers and workers and produced a number that should worry every culture lead in the UK. 80% are concerned their co-workers and teams are using AI to appear more productive than they actually are. The number is not for one cohort. It is the cross-tabulated view, leaders worried about managers, managers worried about workers, workers worried about each other.
The same report puts a second number on the same page. 34% of organisations say their culture is now actively inhibiting their AI transformation goals. A further 65% believe their culture needs to change significantly because of AI. And only 5% say they are making great progress on addressing AI's impact on culture, even though just over half rate the work important or very important.
Read these together and the picture is sharp. Most organisations have rolled AI into the workflow. Almost none have rolled it into the culture. The visible symptom is the 80%. The hidden cause is the 5%.
Why The Trust Number Is Now The Productivity Number
This is not a technology story. It is a behaviour story dressed as one. Deloitte's authors frame it as "culture debt", the negative consequences an organisation accumulates by neglecting culture during transformation. Michael Ehret of Walmart, quoted in the report, puts it more simply: "Those that use technology and AI to advance their mission and values will be better positioned to adapt over time." The implication is straightforward. The companies who will win the AI productivity race are not the ones with the best tools, they are the ones whose people trust each other to use them honestly.
The behavioural research backs that up. Amy Edmondson (1999), in Administrative Science Quarterly, defined psychological safety as the shared belief that the team is safe for interpersonal risk-taking. Where safety is high, people will openly say I used Copilot to draft this or I am not sure I am using this tool well, can we look at it together. Where it is low, they hide the prompt and let the manager guess. Google's Project Aristotle (2015) found psychological safety to be the strongest of the five dynamics that distinguished their highest-performing teams. The 80% trust crack is what a low-safety culture looks like when AI is added to it.
What Most Organisations Try, And Why It Misses
The standard responses to the 80% number tend to fall in three groups. New AI usage policies. Mandatory AI ethics e-learning. A town-hall message asking everyone to "be transparent" about how they use AI. None of these are wrong. All of them are the wrong primary lever.
The reason is the same reason most behaviour interventions miss. Roediger and Karpicke (2006), in Psychological Science, established that active retrieval improves long-term retention by around 50% compared with re-reading or passive review. A 40-minute AI ethics module sits firmly on the wrong side of that curve. A manager who has watched the module has, in retention terms, skimmed a chapter. Three months later, when a team member produces a deck that looks suspiciously polished, the manager has nothing rehearsed to draw on. They guess. They either let it pass and quietly add the person to a mental "probably AI" list, or they ask a clumsy question that lands as an accusation. Either move adds to the debt.
The Conversation Most Managers Have Not Rehearsed
The behaviour that closes the trust crack is a specific kind of conversation, and almost no one has had a chance to practise it. Three observable moves predict the good outcome.
- Open with curiosity, not audit. "Walk me through how you put this together" sits very differently from "did you write this yourself". The first invites the person to share their workflow, including the AI step, without losing face. The second tells them to hide it.
- Name the team norm out loud. The functional teams in the Deloitte data are the ones who have explicitly agreed what "good AI use" looks like, in this team, on this kind of work. A two-sentence team norm beats a fifteen-page policy.
- Reward the disclosure, not just the output. The behaviour most needed is the team member who says "I used the model for the first draft, here is what I changed and why." The behaviour most commonly punished is exactly that, because honest disclosure makes the contribution feel smaller. Until the reward structure flips, the 80% will not move.
None of these moves are in a policy. All of them are observable. None of them transfer from a slide.
What We Do About It
Our leadership and manager workshops and immersive simulations are built for exactly these conversations. Small groups of managers walk into designed scenarios, the team member who hid the AI step, the high performer whose output suddenly doubled, the meeting where someone has to set a team norm in front of their peers. Professional actors play the counterparts. Managers run the conversation, get feedback, run it again. By the third repetition the curious-not-audit move is the default rather than the exception.
In Sidestream's own academic behaviour-change work, building on research from UCL, Cambridge and Bocconi, participants who learned a communication skill through immersive role-play scored roughly 20% higher on observed behaviour than those who learned the same content through video or slide-show training. Self-rated confidence did not predict observed performance. People believed they could have the conversation. They could not, until they had run it three times with feedback. That is the gap between an AI policy on the intranet and a team in which 80% no longer suspect each other.
Culture is not a barrier to AI transformation, it is the operating system AI runs on. The 80% is what happens when you upgrade the apps and leave the operating system unpatched.
The 80% trust crack is not going to close with a new policy. It will close when the conversations about AI use are conversations your managers and teams have had so often they are boring. Read our piece on AI adoption and manager modelling, or book a call to look at what AI-trust rehearsal would look like in your context.
Book a free 30-minute diagnostic call → or read about our research-backed approach.