Fear, Leadership and AI

What this isWat dit is

A research-led essay and practical framework on fear, identity, responsibility and human agency during AI transformation.

ContextContext

A 14-minute research edition in the AI & Human Agency series. It combines organisational psychology, labour evidence, professional observation, Islamic and Jewish ethical sources, an original Courageous Agency Matrix and a six-question leadership diagnostic.

Role:Rol: Author

Format:Formaat: Interactive research essay

ImpactImpact

Reframes fear as organisational information rather than simple resistance and gives leaders a practical framework for creating agency, candour and responsible action during AI transformation.

The fear paradox

Leadership is not the removal of fear. It is the creation of agency while fear is present.

When should leaders reduce fear—and when should they listen to what fear is trying to protect?

AI transformation is also an identity transformation. When expertise, status, control and professional meaning are unsettled, fear may appear as silence, delay, overconfidence, governance theatre or resistance. The leadership task is to turn fear into information, information into shared judgement, and judgement into responsible action.

• Signal — Fear carries information. It may reveal exposure, loss, ethical risk or a threat to dignity, not simply reluctance to change.
• System — Culture shapes fear’s path. Fear becomes silence where candour is punished, and inquiry where questions are protected.
• Agency — Courage is a capacity. People act deliberately when they have voice, boundaries, knowledge and meaningful choices.

The emotion missing from the transformation plan

The strategy meeting ends with agreement. The slides are approved. The organisation is now, officially, “AI-first.” Then almost nothing changes.

Employees experiment privately because they are unsure which tools are permitted. Managers ask for innovation but avoid defining what automation could mean for roles. A pilot produces uncomfortable results, so the lessons are softened before they travel upwards. Governance becomes a reason to pause everything—or a box to tick after a decision has already been made.

I have spent enough time around technology transformation to become interested in the emotion that rarely appears in the programme plan: fear. Not fear as melodrama. Fear as an organisational force—quiet, intelligent, sometimes distorted, sometimes entirely justified.

AI reaches further into professional identity than many previous workplace technologies. It can draft, analyse, recall, recommend, code and converse: activities through which knowledge workers have often demonstrated their value. The question beneath “Will this tool improve productivity?” can therefore be more intimate: If it can do part of what made me useful, what makes me valuable now?

We often call difficult behaviour “resistance.” It is useful in project plans because it turns a complex human response into something measurable and manageable. Yet a finance specialist may be protecting a control nobody else understands; a junior employee may fear that a basic question will expose him as replaceable; a manager may resist a pilot that redistributes expertise and authority; and a legal or security colleague may see obligations an enthusiastic team has missed. Fear is not a diagnosis of another person’s motives. It is a hypothesis to investigate.

Fear is real. The future is not settled.

The ILO’s task-level global index says one in four jobs has some generative-AI exposure, while transformation is more likely than full replacement because most occupations still contain tasks requiring human input.¹

A study of 5,172 customer-support agents found an average productivity gain of 15%, with the largest gains among less experienced workers.² It is strong field evidence, not a universal forecast.

Microsoft reports that 43% of workers globally fear AI may take their job, while only 21% of leaders at selected “Frontier Firms” said the same.³ Corporate research is a useful signal, not neutral truth.

Exposure is not the same as job loss; a productivity result in one workflow does not transfer automatically to every profession; and reported confidence is not proof of good implementation. Mature leadership resists both panic and inevitability.

What fear does to an organisation

Under threat, attention can narrow and people often prefer familiar options, immediate protection and information that confirms an existing view.⁴ But the popular story that an “amygdala hijack” explains organisational behaviour is too simple. Human threat responses depend on context, perceived control, relationships, prior experience and the duration of stress. A boardroom is not a laboratory, and a culture is not a brain.

The organisational question is therefore less “Are people afraid?” than “What becomes possible when fear enters this system?”

Where people expect interpersonal punishment, fear travels towards concealment. Amy Edmondson describes psychological safety as a climate in which interpersonal risk-taking is possible—not comfort, consensus or the absence of standards.⁵ In an AI programme, that includes being able to say: I do not understand this model. The output looks persuasive but wrong. The pilot failed. The customer could be harmed. I used an unapproved tool. I think the executive assumption is mistaken.

An organisation can announce experimentation while punishing every condition experimentation requires. If failure damages reputation, skill gaps affect promotion or dissent is labelled negativity, employees learn the real rule quickly: perform confidence.

This is how fear becomes expensive. Weak signals stay local. Shadow use grows. Risks surface late. Leaders receive cleaner information and make poorer decisions.

Psychological safety in the AI era begins when “I don’t know” can improve a decision instead of diminishing a career.

AI is an identity challenge

Organisations frequently treat AI adoption as a technical transition: select a platform, secure the data, train users and measure uptake. Those tasks matter. They are also incomplete.

Ronald Heifetz’s distinction between technical and adaptive challenges is useful. Technical problems can be addressed with existing expertise. Adaptive challenges require people to revise habits, loyalties, roles or beliefs.⁶ Connecting a model to a knowledge base is technical. Deciding what professional judgement must remain human, whose expertise counts, how accountability changes and what a good job becomes is adaptive. These are not purely cultural questions: under the EU AI Act, certain AI systems used in employment and worker management can fall within the high-risk category.¹³

More training does not automatically resolve fear. Training can improve competence while leaving deeper uncertainty untouched. A lawyer can learn to prompt and still wonder whether careful reasoning will be valued. A seller can use an agent and still suspect that the resulting efficiency target will consume the time saved. A manager can sponsor adoption while sensing that a team with direct access to analysis may need less gatekeeping.

The answer is not to promise that nothing will change. The answer is to make redesign discussable: which tasks may disappear, which decisions remain accountable to humans, what new capabilities will matter, what support exists and where people can influence the outcome.

• The expert: “My value came from knowing.”
• The producer: “My value came from making.”
• The gatekeeper: “My value came from access and translation.”
• The decision-maker: “My value came from choosing.”

What if managers are the ones most afraid?

AI anxiety is often projected downward: employees are imagined as fearful while leadership is cast as rational and future-facing. That picture deserves challenge.

Generative AI can erode informational advantages, accelerate analysis and let employees produce work that once required specialist teams. It may expose how much management depends on coordinating information rather than exercising judgement. It can also create new accountability without providing reliable control. These are legitimate pressures, not proof of bad faith.

But status preservation can disguise itself as strategic caution. The test is not whether a leader slows a deployment. The test is whether the reasons can be examined, whether evidence could change the decision and whether the people affected can participate in defining the risk.

Fear at the top often has more organisational power than fear below. An employee’s fear may produce silence. An executive’s fear can become policy.

Two fear narratives compete for the room

AI discourse often offers leaders two dramatic scripts. The first says: adopt immediately or become irrelevant. The second says: AI may make work, truth or even humanity unrecognisable. Both can collapse judgement into urgency.

The 2025 Microsoft Work Trend Index illustrates the first pressure: 53% of surveyed leaders said productivity must increase while 80% of workers and leaders said they lacked enough time or energy to do their work.³ AI is positioned as capacity “on tap.” Leaders should learn from the data while asking what the framing leaves out: Who receives the gains? Does saved time become recovery, better work or simply a higher target? Which work should not be accelerated?

The opposite narrative can be equally paralysing if every uncertain consequence becomes catastrophe. The OECD’s workplace survey found present-day benefits around job satisfaction, health and wages alongside risks involving privacy, work intensity and bias—and a gap between workers’ current experience and fears about the future.⁷

Good leadership protects the organisation from both manufactured urgency and comfortable denial.

Fear, hope and responsibility

Older traditions ask a more durable question: not “How do we become fearless?” but “How should human beings act when outcomes are uncertain and power carries moral weight?”

In al-Ghazali’s treatment, khawf and raja’—fear and hope—are not slogans or opposites from which one must choose. Each corrects an imbalance in the other. Hope without the means to act becomes wishful thinking; fear without proportion can become despair. His moral psychology is theological, not a management model, but it offers a caution: the right response depends on the condition being treated.⁸

The Qur’anic sequence around tawakkul is striking: show mercy, pardon, consult, decide—then place trust in God. Tawakkul is not passivity or a substitute for due diligence. Consultation, action and trust remain together; uncertainty remains, but responsibility does not disappear.⁹

Amanah can mean trust, obligation or stewardship. Qur’an 33:72 presents it as a grave burden assumed by humanity. Applied cautiously by analogy, technological power is not simply an asset to possess. It is something entrusted, carrying duties toward those who bear its consequences.¹⁰

Pirkei Avot teaches that one is not required to complete the work, yet not free to abandon it.¹¹ This is bounded responsibility: act faithfully without pretending to command every outcome. Rabbi Jonathan Sacks similarly warns that leaders can fail by refusing change or by forcing too much too quickly.¹²

Islamic and Jewish traditions are not interchangeable, and neither exists to validate a corporate framework. The dialogue here is narrower: both contain resources for thinking about action, limits, hope and responsibility without requiring the fantasy of certainty.

What if some resistance is wisdom?

Fear can distort judgement. It can also direct attention toward a real hazard. A person questioning AI may be defending status—or protecting a customer from a failure mode that leadership has not seen.

The distinction cannot be made by tone alone. Responsible dissent is usually specific: it identifies who may be harmed, names an assumption, proposes a test or asks for evidence. Defensive resistance tends to make the claim impossible to examine: the risk remains vague, standards move after each answer or no bounded experiment is acceptable. Yet even this is not a moral verdict. People may lack the safety or language to express a valid concern precisely.

Leaders should therefore treat resistance as data before treating it as obstruction. The same scrutiny must be applied to enthusiastic advocacy: reckless optimism can also protect status, budgets and commercial interests.

• Responsible dissent may name a concrete harm or obligation, expose an untested assumption, invite evidence or a bounded test, change when the evidence changes and represent knowledge close to the work.
• Defensive resistance may keep the threat abstract and total, move acceptance criteria repeatedly, deny any safe route to learning, use governance selectively and centralise voice, access or authority.

The Courageous Agency Matrix

Fear and agency are not opposites. The healthiest zone may contain both: enough fear to perceive what is at stake, and enough agency to respond deliberately.

Courageous Agency Matrix — Daphne Iris van Vliet, 2026.

This is a reflection tool, not a validated psychometric instrument. Use it to begin a conversation, not to label a team.

• Complacency — Low fear, low agency. Drift, dependency and passive adoption. Make the stakes concrete and give people a real decision to influence.
• Paralysis — High fear, low agency. Silence, concealment and symbolic compliance. Reduce interpersonal risk, define safe boundaries and return meaningful choices.
• Experimentation — Low fear, high agency. Learning and speed with possible blind spots. Preserve speed while inviting dissent, red-teaming assumptions and naming who could bear harm.
• Courageous stewardship — High fear, high agency. Candour, scrutiny and courageous action. Protect candour, keep accountability visible and act at a pace the evidence can support.

Five moves that build courageous agency

The aim is not a fearless organisation. It is an organisation capable of noticing fear, interpreting it well and acting without surrendering judgement.

• Surface — Name what may be lost. Discuss roles, status, control, customer impact and professional meaning, not only adoption benefits.
• Separate — Distinguish signal from story. Ask what evidence supports the fear, what remains unknown and who has knowledge closest to the risk.
• Bound — Create safe-to-learn tests. Define scope, human oversight, data boundaries, stopping conditions and what the experiment must teach.
• Share — Redistribute real agency. Let affected people shape workflow design, escalation routes, success criteria and the use of saved capacity.
• Steward — Make responsibility visible. Name who decides, who can challenge, who bears harm and who is accountable when the system is wrong.

Can fear speak here?

For this five-minute leadership reflection, answer never, sometimes or consistently for what people experience—not what policy promises.

A low score suggests fear is likely to travel through silence, private workarounds or compliance. A middle score suggests some foundations exist but employees may still calculate when candour is safe. A high score suggests meaningful voice and accountability; test that conclusion with people who hold less authority.

What is your organisation afraid to say about AI? Ask anonymously first, then examine the conditions that made anonymity necessary.

• People can admit they do not understand an AI system without losing credibility.
• Employees know which decisions remain human and who is accountable.
• A failed pilot can be reported without being rewritten as success.
• People affected by automation help redesign the work.
• Ethical and operational dissent has a protected route to influence a decision.
• Leaders explain how productivity gains will affect workload, roles and rewards.

The opposite of fear is not confidence

Confidence can be performed. It can be borrowed from a persuasive vendor, a charismatic executive or a benchmark that does not fit the organisation. Confidence can accelerate action while weakening judgement.

Agency is different. Agency means understanding enough of the situation to make a choice, having a meaningful role in that choice and accepting responsibility for what follows. It can coexist with uncertainty. It can even coexist with fear.

The goal should not be to move every person from fear to enthusiasm. Enthusiasm is not consent, competence or wisdom. The goal is to create conditions in which people can tell the truth about what they see, learn through bounded action and participate in shaping the systems that will shape them.

There is no responsible route to AI transformation that avoids loss. Some tasks will change. Some expertise will become less scarce. Some roles may disappear; others will emerge; many will be reassembled. Honest leadership does not use hope to deny this, nor fear to force compliance.

It practises something harder: clarity without omniscience, pace without panic, and stewardship without the illusion of control.

If artificial intelligence makes us reconsider what machines can do, fear makes us reconsider what humans believe makes them valuable. The defining question may be whether we can build institutions in which people still have the voice, judgement and courage to remain responsible alongside intelligent systems.

Method and intellectual limits

This is a practitioner essay, not a systematic literature review or theological ruling. It combines organisational research, labour evidence, direct professional observation and carefully bounded interpretation.

The Courageous Agency Matrix and reflection questions are original conceptual tools by Daphne Iris van Vliet. They have not been psychometrically validated. Their purpose is to improve executive conversation and inquiry, not to diagnose individuals, rank teams or replace legal, psychological, security or workforce expertise.

Claims were selected for durability over drama. Corporate research is identified as such; exposure is distinguished from job loss; religious sources are linked directly; and interpretive analogies are labelled.

Sources and limitsBronnen en grenzen

1. Global labour research: ILO — Generative AI and Jobs: A Refined Global Index of Occupational Exposure — 2025; exposure is not observed displacement. ↩
2. Field study: Brynjolfsson, Li and Raymond — Generative AI at Work — Staggered rollout among 5,172 customer-support agents; context-specific evidence. ↩
3. Corporate research — read critically: Microsoft 2025 Work Trend Index — Survey, telemetry and LinkedIn data produced by a major AI vendor. ↩
4. Organisational psychology: Staw, Sandelands and Dutton — Threat-Rigidity Effects in Organizational Behavior: A Multilevel Analysis — Administrative Science Quarterly, 1981; classic account of threat and rigidity. ↩
5. Peer-reviewed review: Edmondson and Bransby — Psychological Safety Comes of Age — Annual Review of Organizational Psychology and Organizational Behavior, 2023. ↩
6. Adaptive leadership: Ronald A. Heifetz — Leadership Without Easy Answers — 1994; technical and adaptive challenge distinction. ↩
7. Cross-country survey: OECD Employment Outlook 2023: Artificial Intelligence and the Labour Market — Workers and employers in manufacturing and finance across seven countries. ↩
8. Classical Islamic ethics: Al-Ghazali — The Book of Fear and Hope — Book 33 of Ihya’ ‘Ulum al-Din, translated by William McKane. ↩
9. Qur’an: Qur’an 3:159 — Mercy, pardon, consultation, decision and trust. ↩
10. Qur’an: Qur’an 33:72 — Amanah has a rich interpretive history; technology as amanah is an ethical analogy. ↩
11. Mishnah: Pirkei Avot 2:16 — Rabbi Tarfon on bounded responsibility. ↩
12. Modern Jewish thought: Rabbi Jonathan Sacks — Pacing Change — Leadership, change, pace and human capacity. ↩
13. Regulatory context: European Commission — AI Act overview — Employment and worker-management AI systems can fall within high-risk categories; details continue to evolve. ↩

Related workGerelateerd werk

NextVolgende

Use the six-question reflection with a leadership team and change one visible behaviour around the lowest-scoring condition.

Back to WritingTerug naar Writing