The AI Sabotage Data Is Real. What We're Doing With It Isn't Helping
Twenty-nine percent of employees admit to sabotaging their company's AI strategy. Seventy-five percent of executives admit the strategy is more for show than actual guidance. The story most people are telling about those numbers misses what the data is describing — and misses where the leverage is.
Writer AI's 2026 enterprise study found that 29% of employees admit to sabotaging their organization's AI strategy — 44% among Gen Z — while 75% of executives concede their AI strategy is more for show than operational. Both numbers describe the same failure. It is not a workforce problem, and it is not a leadership failure. It is a set of missing conditions inside organizations moving faster on AI than they have built the trust to absorb it. Startups and growth-stage companies are most exposed to the pattern. They are also best positioned to solve it.
The Numbers, Read Carefully
Writer AI's 2026 study surfaced a cluster of findings that has been widely cited and, in most coverage, badly misread.
29% of employees admit to sabotaging their company's AI strategy — jumping to 44% among Gen Z. Sabotage in the study is defined broadly: entering proprietary data into public tools, using non-approved software, generating deliberately low-quality outputs, refusing training, tampering with performance metrics to make AI look bad.
76% of executives view employee sabotage as a serious threat. And 75% of executives admit their own AI strategy is more for show than actual guidance.
The dominant reading treats these as separate problems — a Gen Z problem, a workforce readiness problem, an executive credibility problem. Read together, they describe one pattern. A workforce is quietly working around a strategy that its own executives privately consider theater. Both groups are reading the same situation accurately. Neither has a channel to say so out loud.
Sabotage as a Rational Adaptation
What the data describes is not villains. It is people responding sensibly to incentives they can see.
An employee who suspects the AI mandate has been announced ahead of the capability to support it will not raise that concern in an all-hands. They will find a workaround. That workaround, at scale and over time, shows up in the survey as sabotage. What it actually is: unheard feedback that has migrated to the only channel available.
An executive asked by a board or an investor to demonstrate AI-forward posture will publish the strategy on the timeline that pressure requires, not on the timeline the organization can absorb. The gap between announcement and readiness is not bad faith. It is a system running faster than the trust conditions inside it can support.
Both are rational. Both compound.
"A workforce is quietly working around a strategy that its own executives privately consider theater. Both groups are reading the same situation accurately. Neither has a channel to say so out loud."
Why Startups and Growth-Stage Companies Are Most Exposed
This pattern is present across every organization moving fast on AI. It is most acute in startups and growth-stage companies for reasons worth naming directly.
The founder or CEO's public posture is the organization's internal posture. There are no intermediate layers to soften or translate the message. When the leader signals AI-forward to investors, every employee hears the same signal — and the gap between what has been promised externally and what is actually working internally is visible to everyone.
The workforce skews younger, which means the 44% Gen Z sabotage rate applies to a much larger share of the payroll than at an incumbent.
Hiring compression means the people executing the AI work are often the same people who would otherwise flag that the AI work is not going well. There is no separate function to catch the drift.
And investor pressure to demonstrate AI momentum is at an all-time high — creating exactly the conditions in which show-strategy is not just tolerated but reinforced.
The result is a compressed, high-velocity version of the same failure pattern showing up across the RAND and MIT data: 80% of enterprise AI initiatives failing to deliver measurable business value, 95% of generative AI pilots failing to reach production. The startup version fails faster and quieter, and the runway pays for it.
The pattern is geographically concentrated in ways worth naming. Bay Area startups and growth-stage companies → sit closest to the epicenter — the investor pressure to demonstrate AI momentum is highest here, the workforce skews youngest, and the local discourse rewards public AI-forward posture in ways that amplify the show-strategy risk. European growth-stage companies → face a different version of the same pattern, layered with EU AI Act compliance obligations that raise the cost of a failed rollout considerably. In both markets, the leaders who build the trust conditions first move faster on adoption than the ones who lead with the mandate.
Structural Lag, Applied
I describe this condition as Structural Lag — the gap between the pace at which AI capability arrives and the pace at which organizational and leadership systems adapt to hold it. The sabotage data is one of the cleanest indicators of Structural Lag we have. It is what the gap looks like when it becomes behavior.
The instinct in most organizations is to close the gap by pushing harder: more training, more mandates, more tooling. That instinct makes the gap worse. Structural Lag is a trust condition before it is a capability condition, and no amount of additional pressure on a trust deficit repairs it.
What Earned Candor Looks Like Here
Earned Candor is the condition I coach leaders to build — the trust conditions under which people will tell you what is actually happening in time for you to do something about it. It is not something an organization can mandate through policy. It is trust built through specific, observable leadership behavior over time.
Applied to AI adoption in a startup or growth-stage org, it looks concrete. The founder visibly changes their own workflow using the tools the team is being asked to adopt, and talks openly about what is not yet working for them. The strategy update is honest about which parts are proven and which parts are being explored. The first employee who says this tool is not delivering what we hoped is treated as a signal-carrier, not a dissenter. The team learns that reporting real friction is faster than routing around it.
None of this is soft. It is the specific capability that separates the 20% of organizations getting real returns on AI from the 80% still cycling through failed pilots.
What This Looks Like on any given Tuesday
A growth-stage CEO opens her weekly leadership meeting with the two AI experiments she personally ran last week — one that worked, one that didn't. She names both. Her VP of Engineering does the same. By week three, a mid-level PM raises that customer support has quietly reverted to the old workflow because the AI rollout skipped a critical edge case. The problem gets fixed in a sprint instead of surfacing in a Q4 review as a missed OKR.
That is the shift. Not dramatic. Specific. Repeatable. It is the actual leverage point the sabotage data has been pointing at all along.
The Broader Frame
The AI fluency the market is developing is a relational skill, not a technical one. AI moves in and out of precision unpredictably — the same way human beings move in and out of clarity and peak performance. The leaders who integrate AI most effectively bring the same patience to the tool that great managers bring to people. The response to an off moment determines the trajectory more than the off moment itself.
That same patience is what earned candor requires from a leader in relation to a team. Same skill, pointed in two directions.
Startups and growth-stage companies that build this capability while their competitors are still fighting sabotage will not just avoid the failure pattern. They will compound faster than the organizations spending the same period rediscovering why the last mandate didn't take.
Let’s Talk
If the pattern in this piece is one you recognize inside your own organization, I work with senior leaders on exactly this. See AI Executive Coaching → for individual work and Organizational AI Coaching → for team engagements.
Sources
Writer AI, AI Adoption in the Enterprise 2026
MIT NANDA, The GenAI Divide: State of AI in Business 2025
RAND Corporation, Enterprise AI Initiative Failure Analysis
Mercer, 2026 CEO Survey on AI-Driven Workforce Change