AI Is Anti-Staff by Default — and Staff Are Anti-AI by Default

We need an HR seat on AI governance. Not as a courtesy. As a structural requirement.

The Graphic Everyone Shares — and What It’s Missing

There’s a governance graphic that’s been circulating on LinkedIn. You’ve probably seen it. Two columns: Data Governance (trustworthy inputs) on the left. AI Governance (trustworthy outcomes) on the right. It’s clean. It’s sensible. It covers bias, explainability, robustness, accountability.

It’s also structurally incomplete.

The graphic asks “trustworthy outcomes” and answers: for customers, for regulators, for the business. But it never asks: trustworthy outcomes for the staff?

That’s not a minor omission. It’s the gap that kills rollouts.

Because 80% of AI projects fail — twice the rate of non-AI IT projects — and the gap is largely governance and organisational, not technical.¹ Cultural challenges and change management are the primary obstacle to AI adoption, with 91% of data leaders pointing to people issues rather than technology challenges.²

The model works fine. The organisation falls apart around it.

AI Is Anti-Staff by Default

This isn’t a moral claim. It’s a mechanical one.

AI collapses the cost of cognition. Tasks that took hours take minutes. Throughput that required a team can be handled by one person with an AI assistant. That’s the pitch. That’s the business case.

But here’s what happens next — not sometimes, reliably:

A UC Berkeley research team followed 200 workers at a tech company for eight months after AI was introduced. They found that workers “worked longer hours at a faster pace on a broader range of tasks.” Not because they were told to. Because the system created implicit pressure. AI didn’t reduce work — it intensified it.^3,4

This is what I call the extraction reflex. AI lifts throughput, and management (often unconsciously) does what management always does with productivity gains:

• “You can handle more cases now.”
• “We’ll hold headcount flat next cycle.”
• “We’ll measure you more tightly because we can.”
• “We’ll hit the same SLA with fewer people.”

Even when the work becomes less clicky, it becomes more relentless, more monitored, more standardised, and less autonomous.

The evidence is stacking up

Burnout: 83% of workers report some degree of burnout. The influence of burnout on engagement jumped from 34% to 52% in a single year.⁵ Mental fatigue and cognitive strain are now the leading indicators of burnout — surpassing workload volume for the first time.⁵

Surveillance: 74% of U.S. employers use online monitoring tools. 61% use AI-powered analytics to measure employee productivity or behaviour.⁶ Employees in high-surveillance workplaces report 45% stress levels compared to 28% in low-surveillance environments. Eight in ten say monitoring erodes trust.^7,8

Deskilling: AI coding assistants deliver minimal productivity gains while causing a 17% reduction in skill development.⁹ Because AI handles the messy, repetitive tasks that once built judgment, junior employees miss chances to develop it — and organisations risk ending up with managers who’ve never done the underlying work.¹⁰

Role erosion: AI strips out judgment work, leaving humans with exception handling and blame. The work doesn’t disappear — it concentrates into the hardest, most emotionally draining parts.¹¹

“What if you speed up someone’s workflow by 50% or 100%? Are they now expected to work on two or three times as many cases in the same day? What does that do to staff morale?”

This isn’t a side effect. It’s the default trajectory. AI chews up people if you don’t take care of it.

Staff Are Anti-AI by Default

And here’s the symmetry that most governance frameworks miss entirely.

If AI is structurally extractive toward staff, then staff resistance to AI isn’t irrational — it’s the correct response to misaligned incentives.

But sabotage is just the visible tip. The full spectrum of rational resistance includes:

• Low-quality requirements: “Don’t tell them the real workflow”
• Adversarial testing: “Prove it fails”
• Passive non-adoption: “Sure, we tried it”
• Shadow resistance: comply outwardly, route around it quietly

The trust numbers are devastating. Trust in company-provided generative AI fell 31% between May and July 2025. Trust in agentic AI — systems that act independently — dropped 89% in the same period.¹³

Meanwhile, 64% of managers say employees fear AI will make them less valuable. 58% say employees fear it will cost them their jobs.¹⁴ And 56% of employees say they’ve received no AI training or minimal mentorship.¹⁵

Executives believe their workforce is informed and enthusiastic about AI. They’re wrong. Most employees report confusion, anxiety, and limited involvement in key decisions.¹⁶

“Most staff are anti-AI anyway. The general consensus from employees is: it’s taking jobs, and there’s no real upside for them.”

That’s not a character flaw. That’s incentives plus fear. And it’s entirely predictable.

Most organisations accidentally send this message: “We’re introducing AI to increase productivity. If it works, we’ll raise targets or cut roles. If it breaks, you’ll wear the blame.” That is a perfect recipe for resistance.

The Governance Gap: We Govern the Model but Not the Humans

Look at any standard AI governance framework. NIST AI Risk Management Framework defines seven characteristics of trustworthy AI: validity, safety, security, accountability, explainability, privacy, and fairness.¹⁷

All focused on the model. None address what happens to the people the model replaces, monitors, speeds up, or deskills.

Look at who sits on AI governance committees. Harvard Law research shows 72% of boards involve CTO/CIO in AI discussions. 27% involve CFO. HR? Available “on demand” — not a permanent seat.^18,19

This is a structural problem, not an oversight. AI governance emerged from data science and ML ops communities. They think about model behaviour. HR involvement feels “soft” and “political.” So HR gets excluded from the governance table — and workforce impact becomes ungoverned.

80% of AI project failures are organisational, not technical.¹ Yet governance frameworks focus on the 20% (model risk) and ignore the 80% (workforce and organisational risk).

AI governance that ignores the workforce dimension is like building a bridge that’s structurally sound — but points it at a cliff. The engineering’s fine. The outcome is still a disaster.

The Third Column: Workforce Governance

The fix isn’t more change management comms. It’s structural.

If Data Governance covers trustworthy inputs and AI Governance covers trustworthy outcomes, then the missing column is Workforce Governance: trustworthy work.

This isn’t “being nice to staff.” This is governance — with the same rigour we apply to data quality and model safety. Enforceable policies. Measurable risks. Clear ownership. Defined triggers.

We already apply “Architecture, Not Vibes” to AI systems — we don’t trust AI to behave, we architect containment. The same principle applies to AI’s impact on workers. Structure beats intentions.

The Workforce AI Compact

What does workforce governance look like in practice? Not a values statement. Not a slide deck. A governance instrument — pre-negotiated, enforceable, with clear triggers and stakeholder accountability.

Here are the five policy elements:

1. Throughput Policy

For the first 6-12 months of any AI deployment, productivity gains are used to buy time back, reduce backlogs, improve quality, and train — not to ratchet quotas.

The principle: AI should buy time before it buys headcount.

This isn’t permanent protection from change. It’s a governed transition window. EY’s 2025 research shows that among AI-productive organisations, 83% reinvested gains rather than cutting headcount — and they outperformed the 17% who went straight to extraction.²⁰

2. Value-Sharing Policy

If AI lifts productivity, staff share in the benefits: pay, time, flexibility, progression pathways. Not all of it. But a defined share, governed by policy.

Industries where AI gains were shared saw wages rise twice as fast as those where they weren’t.²¹ The productivity-pay gap has been widening since the 1970s — AI will accelerate this unless governance intervenes.²²

3. No-Surveillance-Creep Rule

AI can assist work. It can’t quietly become a panopticon without explicit governance approval. Any expansion of monitoring scope, frequency, or granularity requires sign-off from the governance committee — with HR at the table.

Currently 74% of employers monitor, 61% use AI analytics, and 59% of employees report surveillance-induced stress.^6,7 This is ungoverned expansion. The Compact makes it governed.

4. Role Horizon Plan

For every AI deployment: what happens to affected roles at 6, 12, and 24 months? Redeploy, reskill, transition — and what budget backs that promise?

55% of employers who laid off workers for AI now report regretting it.²³ Half of AI-attributed layoffs will be quietly rehired, often offshore or at lower salaries.²³ A role horizon plan prevents knee-jerk decisions that damage both staff and capability.

5. Accountability Clarity

When something goes wrong, who owns the decision? Not the frontline worker who clicked “approve.” Humans aren’t scapegoats for model errors. Escalation pathways and decision rights are explicit, not implied.

Without this, you get “accountability without authority” — humans blamed for AI mistakes they couldn’t prevent and weren’t empowered to override.

HR as the Staff Champion

HR can’t just be “change management comms.” HR needs a formal, permanent seat inside AI governance.

Not because HR understands model architecture — that’s not their job. But because HR owns the systems that AI disrupts: rewards, performance, role design, capability, and trust. HR is the natural steward of workforce governance, the same way Legal stewards regulatory compliance without writing code.

Right now, HR is “on demand” in most governance structures.¹⁹ That’s like having Legal available “when the topic comes up.” By the time they’re called in, the decisions are already made.

What HR’s governance role looks like:

• Pre-deployment: Role impact analysis, throughput policy sign-off, value-sharing terms agreed
• During deployment: Monitor morale, burnout, adoption quality (not just adoption rate), escalation paths for workforce concerns
• Post-deployment: Role horizon execution, skill development tracking, Compact compliance review
• Veto power: On deployments that violate the Workforce AI Compact — the same way Finance can veto projects that exceed risk appetite

This turns staff from “test subjects” into “co-owners.” When employees know there’s a governed Compact with real protections, requirements gathering becomes collaborative instead of adversarial. Staff become your best red-teamers — pointing out failures to harden the system, not to kill it.

The Regulatory Signal Is Already Here

If you think workforce governance is optional, look at what’s already happening in legislation:

• Colorado AI Act (effective Feb 2026): Requires impact assessments, worker notification when AI is used in employment decisions, right to appeal AI decisions, public disclosure of AI systems in use.²⁴
• California: Prohibits discriminatory automated-decision systems in employment, requires meaningful human oversight with trained overriders, mandates proactive bias testing and four-year record retention.²⁵
• Illinois (effective Jan 2026): All employers must notify employees when AI is used in employment decisions.²⁶
• EU: 42.3% of workers already affected by algorithmic management.²⁷

These aren’t change management guidelines. They’re enforceable governance with audit requirements, appeal rights, and penalties. The Workforce AI Compact puts you ahead of the curve — building voluntarily what regulation will eventually require.

The Choice: Govern It or Get Governed

Here’s the bottom line.

85% of enterprises plan agentic AI deployments. Only 21% have governance frameworks ready — and those frameworks don’t cover workforce impact.²⁸

You have two options:

“If you don’t govern the human impact, the humans will govern your rollout for you.”

That’s not ideology. That’s physics.

We’ve got to look after the people in the process. Not because it’s nice. Because AI governance without workforce governance is structurally incomplete — and structurally incomplete governance produces structurally predictable failures.

Add the third column. Give HR a permanent seat. Sign the Compact before you ship the model.

Your AI might be ready. Your governance isn’t — until your people are protected.