What happens to your career if you refuse to use AI at work depends on your job, manager and results: you can remain valuable without using every tool, but you may face fewer promotions, tougher productivity comparisons and a shrinking pool of suitable employers. That is more precise than the headline claim that refusing artificial intelligence means inevitable unemployment.
In 2026, the tension is real. Enterprise AI vendor WRITER reports that 60% of surveyed executives plan to lay off workers who will not adopt AI; meanwhile, research from Duke University finds that employees can be judged more harshly for using AI. Both cannot be understood through a simple “adopt or lose” slogan. One measures executive intentions in a vendor-sponsored survey; the other examines social evaluation in controlled experiments. Neither predicts the fate of an individual employee.
The meaningful question is not whether you embrace or reject technology as an identity. It is where AI materially improves your role, where it adds risk or rework, and whether you can prove the value of the choices you make. A worker who refuses an unapproved chatbot for confidential client material is making a different decision from a worker who rejects a safe internal summarisation tool because learning it feels inconvenient.
This analysis develops two connected ideas: the AI-refusal career tax, meaning the extra proof and adaptation burden on non-users; and the AI-dependence tax, meaning the cognitive, reputational and quality costs of handing too much responsibility to models. Neither is a measured economic index. Both are explanatory frameworks for making better decisions about work.
The Career Tax: What Refusal Changes Before Anyone Is Fired
An AI-refusal career tax develops when an employer begins assessing output against a workflow that assumes some automation. The “tax” can take the form of extra hours needed to match throughput, more justification for a slower approach, or exclusion from projects designed around AI-assisted production. It is not a standard payroll deduction and cannot be expressed as a universal percentage.
Imagine two analysts asked to summarise a collection of non-sensitive internal reports. One uses an approved system to assemble an initial thematic map and then verifies every claim; the other reads and organises the same documents manually. If the second analyst identifies crucial inconsistencies that the model misses, their method might produce higher value. If both reach the same defensible answer and the first finishes considerably earlier, the non-user must explain what extra benefit justified the effort.
The Productivity Comparison
Managers compare outcomes more readily than invisible process. In writing, support, basic research and reporting, turnaround time is visible and can become an informal performance measure. That matters most when teams lack independent measures of accuracy, customer confidence or risk reduction. The baseline shifts even when the job description does not.
As how AI reshapes exposed jobs makes clear, technical exposure differs considerably across occupations and tasks. That is why a productivity comparison should be conducted task by task, rather than applied to an entire career. A model may draft routine text efficiently and still be unsuitable for sensitive decisions, rare exceptions or legally accountable recommendations.
The Opportunity Filter
An employer may not explicitly reject applicants who avoid AI. Instead, job descriptions, project staffing and promotion rubrics increasingly mention tools, automation, agents or workflow redesign. A strong specialist can remain employed but lose access to teams where the expected contribution includes organising AI-enabled processes.
The more consequential effect can be cumulative. Missing one cross-functional programme means less leadership exposure; less exposure yields fewer examples of modern transformation work; and fewer examples can weaken the next interview. This is a plausible career pathway, not a quantified causal finding. Workers should track actual selection criteria rather than assume all employers behave alike.
Why Manager Behaviour Can Matter More Than Company Policy
Organisation-wide policy tells employees what tools are approved; a direct manager often decides which behaviours are admired. That distinction helps explain why refusing AI can be neutral in one department and career-limiting in another. It also explains why adopting AI is not always rewarded.
In research published in Proceedings of the National Academy of Sciences in May 2025, Jessica Reif, Richard Larrick and Jack Soll ran four preregistered experiments with 4,439 participants. They identified negative competence and motivation judgements associated with AI assistance. The experiments tested perceptions under defined conditions, not realised long-term career outcomes.
The Duke Paradox
Duke’s account of the hiring experiment described an important split: evaluators who seldom used AI penalised daily AI users, while frequent AI users preferred candidates who also used AI regularly. This is not proof that every manager is biased. It shows a mechanism by which the same candidate might be favoured or rejected depending on the evaluator’s own relationship with the tool.
As researcher Jessica Reif observed in Duke’s 2025 coverage, “People anticipate being judged harshly for using AI.” That concern complicates simplistic advice to display AI use everywhere. A candidate must learn whether the role prizes automation fluency, confidential judgement, speed, original work, or a mix of all four.
Signals to Look for in Your Own Team
Ask what the manager rewards in practice. Are team members promoted for better customer outcomes or for sheer volume? Are errors attributed to an individual even when AI generated the draft? Do leaders publicly share their own AI workflows and verification practices? Are employees encouraged to reject AI output when it is wrong?
An employee should request specific, written expectations: which tasks require approved AI tools, which forbid them, what verification is necessary and what counts as satisfactory performance. This turns a philosophical disagreement into a testable operating standard. It also surfaces differences between permitted AI use and genuinely required AI use.
Where a Deliberate No Can Be a Professional Advantage
There are legitimate cases where refusing a particular model or workflow protects a career. When the cost of a mistaken claim, inappropriate disclosure or misplaced delegation is high, slower human-led handling can create more value than a fast draft. The key is rejecting a risk, not announcing a blanket opposition to technology.
Examples include confidential negotiations, sensitive personnel decisions, privileged legal material, unresolved medical questions and safety-critical judgements. Each comes with different professional and legal duties. A general-purpose AI assistant may still help with approved non-sensitive preparation, but responsibility for decisions remains with the authorised professional.
Trust, Confidentiality and Accountability
Confidentiality concerns deserve their own conversation. A worker who declines to paste client documents into an unapproved chatbot is acting differently from someone who refuses a company-authorised enterprise system with documented controls. Employers need to articulate retention, access, training and incident-reporting rules; employees need to follow them.
An associated article on human work and AI substitution examines the difference between automating components and replacing a professional role. That difference is especially significant when clients pay for independent accountability, not just a deliverable. Where a trusted adviser’s judgement is the product, indiscriminate delegation can damage the service itself.
Craft, Learning and Original Judgement
Creative work also needs nuance. Generative AI can be useful for proofreading, pattern spotting and exploring alternatives; it can be less useful when the customer is buying an individual’s voice, reasoning, lived experience or taste. The reason to withhold AI from a final recommendation should be explained in terms of quality and ownership.
Similarly, junior workers sometimes need to perform fundamentals unaided in order to learn them. Using a model to challenge a first attempt can build capability; having it complete every first attempt can prevent practice. This is a task design issue, not proof that AI inevitably destroys skills. A selective refusal can therefore be an investment in future independence.
Where Total Refusal Becomes a Growing Liability
The career calculus changes when a large share of work is digital, repeatable and relatively easy to verify. A coordinator producing meeting notes, a support agent drafting standard replies or a junior analyst organising routine evidence may face increasing pressure to understand safe automation. When the employer has already redesigned workflows around an approved tool, total non-participation can make someone harder to staff.
The International Labour Organization’s 2025 exposure research stresses that work is made of tasks and that exposure should not be mistaken for elimination. Nonetheless, workers whose responsibilities overlap strongly with repeatable information processing may have to differentiate through escalation handling, contextual knowledge, exception management and quality control.
Routine Digital Production
The least defensible stance is refusing without conducting a credible comparison. A worker should establish what the AI workflow actually does, count review and correction time, examine error severity, and identify any data restrictions. If manual work genuinely performs better on relevant metrics, a documented case is persuasive. If it does not, the person must decide whether values or preferences justify accepting the career trade-off.
Study evidence on AI job displacement before concluding that a particular occupation is safe or doomed. The important variables include task composition, employer investment, customer tolerance for error and the organisation’s ability to redesign responsibilities. A job can change dramatically without disappearing, and new work can emerge alongside displaced tasks.
Hybrid Roles and Recruitment
In recruitment, refusal also changes how candidates describe themselves. Employers may seek someone who can supervise models, evaluate output, document constraints or train teams, even if that person is not a power user. Avoiding daily AI use is compatible with AI literacy; refusing to understand how a tool works is a broader professional limitation.
The magazine’s coverage of AI skills employers actually value shows the distinction between knowing prompts and producing accountable business results. Jobseekers should provide examples of error detection, secure workflow design and independent subject knowledge rather than declaring themselves either “AI-first” or “AI-free”. Those labels are too broad to establish competence.
| Work category | Reason refusal may help | Reason refusal may hurt |
| High-trust advisory | Protects accountable human judgement | May reduce preparation speed |
| Regulated or confidential work | Avoids unauthorised data exposure | May block approved safe workflows |
| Routine digital production | Can preserve specialist quality | Higher throughput expectations |
| Early-career learning | Protects foundational practice | Misses useful feedback and tooling fluency |
What Employer Surveys Actually Prove—and What They Do Not
WRITER’s 2026 enterprise AI survey, developed with Workplace Intelligence, questioned 1,200 executives and 1,200 employees. It reports that 60% of executives plan to lay off workers who will not adopt AI. Published coverage also reports that 77% warned resistant employees about lost advancement and 79% characterised non-adoption as career-limiting.
These are important statements of managerial intention. They do not establish that 60% of companies have already dismissed non-users, or that 60% of all employees face redundancy. Respondents may express ambition, organisational pressure or plans that change with budgets and operational realities. Because WRITER sells enterprise AI technology, readers should also distinguish its survey results from neutral labour-market administrative data.
Read Intentions, Not Predictions
Contradictions within the same survey are revealing. WRITER reports that while 97% of employees derive personal benefit from AI, only 23% of organisations see significant return on investment. That gap suggests workplace pressure can increase before the business case has been thoroughly demonstrated. Employees are entitled to ask for outcomes rather than slogans.
Microsoft’s 2025 Work Trend Index reported that 82% of leaders saw the year as pivotal for rethinking operations and that leaders anticipated growing responsibility for managing AI agents. This supports the claim that many executives expect redesign, but it remains a snapshot of respondents’ expectations, not proof of universal worker outcomes.
Compare Signals Across Sources
The ILO provides the counterweight. Its occupation-exposure work distinguishes technology’s capacity to transform tasks from actual losses of employment. The correct synthesis is that employers may reward adoption sooner than national statistics can establish the long-term effects. Career planning therefore requires attention to both local workplace signals and wider evidence.
An individual should monitor revised role descriptions, training access, evaluation criteria, approved-tool policies and changes in staffing. Those are stronger near-term indicators than a generic prediction about “the future of jobs”. In the UK, UK demand for AI skills gives a useful labour-market perspective on demand for AI-adjacent ability without assuming every worker needs to become an engineer.
| Source | What it measures | What it cannot establish |
| WRITER 2026 survey | Executive and employee views | Actual proportion of future layoffs |
| Duke 2025 experiments | AI-related evaluation biases | Long-term career causation |
| ILO 2025/2026 | Task and occupation exposure | Individual redundancy date |
| Microsoft 2025 index | Leadership plans and work trends | Universal productivity gains |
The Other Career Tax: Becoming Dependent on AI
A worker who refuses AI can fall behind. A worker who delegates judgement uncritically can lose confidence, ownership and trust. Both outcomes are possible, and both should shape a serious career strategy. The appropriate distinction is not human versus machine but who performs the thinking, who checks the evidence and who owns the consequences.
In March 2026, Elena Hayoung Lee, Yidan Yin, Nan Jia and Cheryl J. Wakslak published a Scientific Reports study combining a preregistered experiment of 269 participants with a follow-up survey of 270. Participants using AI by copying generated writing reported reduced independent self-efficacy, ownership and meaning relative to people who wrote themselves or used AI to refine their own drafts. The study concerns measured task experiences and associations, not an inevitable long-term neurological decline.
Passive Use Versus Active Collaboration
Active collaboration produced a more encouraging pattern: participants first performed the substantive work and then used AI to improve it. They maintained a stronger connection to the output. That suggests a practical career rule: let a model question, reorganise and test your thinking, but keep enough first-hand work to understand the result and defend it.
The article on the risk of losing writing skills explores this distinction for writing skills. The same logic extends cautiously to analysis and decision-making: people need opportunities to practise core reasoning without outsourcing every first attempt. Where competence must be demonstrated unaided, use AI as feedback after a genuine attempt.
The Rework and Credibility Cost
Poor AI use can also consume time invisibly. A convincing but inaccurate summary may take longer to check than a short manual reading. Generated code might introduce edge-case bugs; customer messages can sound polished but miss essential context. Productivity should include investigation, correction, approval and downstream consequences, not only the minutes saved producing a first draft.
For content work, using AI without losing quality recommends evaluating the time to a defensible finished outcome rather than raw generation speed. That is a relevant wider lesson: refuse automation that fails a net-value test, but do not treat the first failure as a permanent verdict on an entire class of tools.
A Three-Zone Map for Deciding When to Use AI
A useful decision framework begins with the function of the work, not the brand of tool. The three-zone map below is editorial guidance, not a validated psychometric instrument. It helps professionals classify tasks by responsibility, repeatability and cost of failure.
For a single role, several zones can coexist. A client adviser might use AI to organise public background material, avoid it for a sensitive negotiation and rely on automation for appointment scheduling. Declaring the whole occupation AI-compatible or AI-incompatible conceals those differences.
Zone One: Human Signature
The human-signature zone covers work where trust, originality, fiduciary responsibility, safety or interpretation are central. The safest position may be to use approved tools only around the edges, and never permit a model to make the final judgement. Evidence of advantage should include client confidence, low error rates and defensible decisions.
An employee should write down the tasks in this zone and articulate why the human contribution cannot be delegated. “I refuse AI” is an uninformative statement; “I sign the advice and therefore verify the source documents myself” is a professional rationale.
Zone Two: Augmentation and Zone Three: Scale
The augmentation zone covers research organisation, drafting, code suggestions, structured summarisation and routine data preparation. Here the worker should pilot approved tools, compare output and retain review responsibility. A permanent refusal may impose a growing throughput disadvantage when the tool is reliable and verification remains affordable.
The scale zone concerns repetitive, standardised production where speed and volume are central. In these workflows, either learning safe automation or moving towards exception handling, quality assurance and process ownership may be necessary. A decision to stay entirely manual should be conscious of the likely staffing and opportunity trade-offs.
| Zone | Typical tasks | Practical stance | Evidence to track |
| Human signature | Sensitive advice, safety, original voice | Human final decision; narrow approved support | Trust, errors, accountability |
| Augmentation | Drafting, coding suggestions, research | Trial approved AI with active review | Net time, quality, correction cost |
| Scale | Standardised reporting and support | Automate safely or own exceptions | Throughput, escalation, customer impact |
How to Negotiate an AI Boundary Without Damaging Trust
A principled workplace boundary is more credible when framed as a performance and risk decision. Saying “AI is unethical” may express a sincere value, but it leaves a manager unsure how work will get done. Saying “I will not submit restricted client information to a third-party service; I can use the approved internal system for nonsensitive preparation” offers an actionable alternative.
Begin by separating prohibited use from personal preference. Ask for the authorised tools, acceptable data categories, audit trail expectations and escalation contacts. If the employer requires a tool, understand whether the requirement is written into duties or performance metrics. Do not assume that disagreement gives an automatic legal right to ignore an instruction.
A Better Conversation With Your Manager
An effective script is: “I can meet the accuracy and turnaround standards. I would like to keep final client recommendations human-led because I remain accountable for the evidence. For initial sorting, I am willing to test our approved system and report whether it reduces total review time.” This describes a boundary and an experiment.
A second script is appropriate when results have been poor: “In the last sample, the AI workflow required substantial fact correction. I suggest comparing ten representative cases using total handling time and material-error rate before we make it compulsory.” A controlled pilot is more persuasive than a general claim that models are unreliable.
Record the Trade-Off
Document the agreed approach, including tasks, baseline metrics, conditions for review and handling of confidential information. An employee with a disability, union agreement, contractual protection or another legally relevant concern may have additional rights; the specifics depend on jurisdiction. Seek qualified guidance rather than treating a general article as legal advice.
Do not secretly use an unapproved tool to keep pace. That exposes confidential material and can break governance rules even when the output appears helpful. Nor should you misrepresent human work as AI-driven or AI output as independently researched. Trust is part of the career calculation, not a separate moral accessory.
A 30-Day Career-Proof Experiment
A deliberate experiment can replace debate with evidence. Choose three frequent tasks: one where AI seems clearly useful, one where it seems risky, and one where you are uncertain. Ensure any test is authorised and uses only information permitted by company policy. Do not upload confidential data into consumer accounts to conduct an informal comparison.
Establish a manual baseline using several comparable cases, noting time to completion, correctness, amount of revision and quality of the ultimate outcome. Do not cherry-pick unusually easy or difficult cases. Account for human review, tool setup and any need to correct errors.
Week One and Two: Measure
In week two, trial the approved AI-supported process on comparable material. Identify which part of the task the model performed. A generated outline is not the same intervention as an unreviewed final decision. Record false claims, omissions, privacy concerns, accessibility issues and the number of human corrections.
Compare total process time rather than typing speed. A draft created in one minute but reviewed for forty may be inferior to a thirty-minute careful manual workflow. Conversely, a model that reduces administrative preparation while preserving quality may deserve adoption even if the worker dislikes the initial learning curve.
Week Three and Four: Decide
By week three, discuss results with a manager or professional peer. Propose a division of labour: which steps remain human, which can be augmented, which should be automated, and where mandatory review gates sit. Ask whether this division satisfies performance and compliance expectations.
In week four, write a one-page career strategy: skills you must practise independently, workflows you will learn, tasks you will not delegate and the evidence you will use in promotion conversations. Set a quarterly review. This preserves professional choice without pretending that technology or management expectations will stand still.
| Metric | Manual baseline | AI-supported pilot | Decision rule |
| Total minutes | Track full process | Include correction time | Prefer reliable net savings |
| Material errors | Count verified errors | Count verified errors | Reject unacceptable risk |
| Data handling | Check authorised access | Check vendor and policy | No prohibited disclosures |
| Learning value | Can reproduce unaided? | Can explain and defend? | Preserve core competence |
Promotion, Pay and Job Security: Three Different Risks
Career discussions frequently combine three different outcomes: being fired, losing a promotion and earning less. They require separate evidence. An employee may remain secure in a specialist job yet struggle to demonstrate readiness for a cross-functional leadership role built around modernised workflows.
Promotion criteria are especially sensitive to local management assumptions. A supervisor who equates visible automation with initiative can undervalue cautious work; a supervisor who values judgment may reward careful refusal. The best safeguard is to translate your approach into results that matter across teams: fewer errors, stronger retention, faster resolution of exceptions and verified commercial impact.
Promotion and Role Mobility
Compensation claims deserve restraint. Reports of advertised wage premiums for AI skills are not evidence that each worker receives a raise upon learning prompting. Skills demand varies by occupation, geography and employer. A specialist who helps govern deployment may have valuable AI-adjacent capability even when they do not use models for every daily task.
At interview or appraisal, avoid positioning yourself as someone who will never change. Instead demonstrate ability to evaluate a system’s limits, explain safeguards and deliver measurable work. Employers buy reliable outcomes and adaptable judgement, not adherence to a software ideology.
Pay and Redundancy
Redundancy risk is real in some organisations, but surveys of future plans do not identify whose roles will disappear. Staff cuts may be driven by budgets, restructuring, customer demand or technology together. If a team is reorganised around agents, a fully manual specialist could become harder to redeploy, yet skilled reviewers and escalation experts may become more necessary.
The defensible conclusion is conditional: the more of a role’s value comes from standardised high-volume output, the higher the potential career cost of blanket refusal; the more it comes from trusted judgement and hard-to-automate exceptions, the more plausible selective use becomes. No single industry label substitutes for task-level analysis.
Professional Sovereignty: The Alternative to Both Extremes
Professional sovereignty is the capacity to decide deliberately where AI belongs in your work while remaining responsible for professional outcomes. It is not immunity from an employer’s lawful instructions. It is a disciplined form of agency: understanding the technology sufficiently to set defensible boundaries instead of rejecting it blindly or handing it everything.
Employees who practise it can articulate three kinds of decisions. “I will use AI for administrative preparation when it is permitted.” “I will use it to challenge my own draft, not to replace first-principles reasoning.” “I will not delegate this judgement because I cannot adequately verify the model and I remain accountable to the client.”
Choose Constraints You Can Explain
Such statements are valuable because they survive product change. Specific chatbot brands, pricing tiers and benchmark positions will shift. The underlying questions about data, accuracy, reviewability, responsibility and human learning remain durable. They also travel across companies and sectors.
Professional sovereignty has a cost: you may need to spend more time explaining your choices, and some employers will not accommodate them. Calling that trade-off a values-based choice is legitimate. Pretending that it carries no consequence is not.
Maintain the Right to Reconsider
It also protects people against pressure to appear technologically enthusiastic without evidence. The manager who demands AI for every task should explain the expected gains and risk controls; the employee who refuses every task should explain the corresponding value. Both claims are subject to measurement.
The strongest career position is flexible selectivity. Learn enough to understand AI use by colleagues, preserve skills you cannot afford to lose, stay within workplace policy and judge tools by actual outcomes. Career resilience comes from repeatable professional judgement, not from a permanent yes or no.
Our Editorial Verification Process
This analysis cross-checked the 2026 WRITER enterprise AI adoption report against publisher and secondary reporting; examined the 2025 Duke/PNAS social-evaluation experiments, the 2026 Scientific Reports study of passive and active AI use, the International Labour Organization’s exposure research, and Microsoft’s 2025 Work Trend Index. It distinguished survey intentions from realised employment outcomes, experiments from long-term evidence, and task exposure from layoffs.
This article was researched and drafted with AI assistance for the Awais Khalid editorial desk at Perplexity AI Magazine. Independent human editorial review, primary-source rechecking and final approval remain required before publication. No unperformed verification is represented as completed.
Conclusion
Refusing artificial intelligence at work does not automatically end a career. It changes the case a professional must make for their value, particularly where approved AI systems already shape throughput, collaboration and promotion expectations. The effects are not uniform: a worker protecting sensitive information may be exercising sound judgement, while one refusing to examine an efficient, secure workflow may be giving up opportunities unnecessarily.
The available evidence cuts in both directions. Executive surveys indicate mounting pressure to adopt AI; peer-reviewed work documents stigma against AI users and potential psychological costs when people rely on generated output too passively. Those findings describe different mechanisms. Neither supports universal promises about job security or automatic career failure.
For employers, the unresolved question is whether they can build reliable systems that reward real quality rather than tool usage as a performance signal. For workers, the durable response is selective competence: understand the tools, insist on appropriate controls, preserve core skills, compare the results and explain decisions in professional terms. Refusal becomes most costly when it is unexamined. Adoption becomes most dangerous when it is uncritical.
Frequently Asked Questions
What happens to your career if you refuse to use AI at work?
You may face slower progression, fewer suitable roles or closer productivity scrutiny, especially in AI-intensive teams, but refusal does not automatically mean dismissal. Outcomes depend on duties, employer policy, manager expectations and the quality of your work.
Can my employer force me to use AI at work?
Employers can often set lawful tools and workflow requirements, subject to local law, contracts, collective agreements, accessibility needs and data-protection obligations. Check formal policies and obtain jurisdiction-specific advice where necessary.
Will refusing AI stop me getting promoted?
It can if AI-related workflow capability is a documented promotion criterion or valued by your manager. Demonstrating strong results, safe evaluation skills and readiness to learn may reduce the risk.
Will I lose my job if I do not use AI?
No general evidence establishes that an individual will lose their job solely because they do not use AI. Some executives say they plan workforce changes, but survey intentions are not actual dismissal rates.
Is it better to use AI secretly?
No. Unapproved AI use may expose confidential information, breach employer policy and damage trust. Request approved tools and clear boundaries instead.
What if AI output is worse than mine?
Measure comparable work, including correction time and severity of errors. If human-only work produces better overall results, explain and document that advantage rather than relying on preference alone.
Should I learn AI if I do not want to use it daily?
Usually yes. AI literacy helps you assess coworkers’ outputs, protect data, question unreliable results and make informed career choices without making AI the centre of every task.
Which jobs can remain more human-led?
Roles built around accountable judgement, trust, complex physical work, sensitive decisions, relationships and original craft may preserve stronger reasons for human-led processes. Even these jobs can include tasks suitable for approved assistance.
References
Reif, J. A., Larrick, R. P., & Soll, J. B. (2025). Evidence of a social evaluation penalty for using AI. Proceedings of the National Academy of Sciences, 122(19), e2426766122. https://doi.org/10.1073/pnas.2426766122
Lee, E. H., Yin, Y., Jia, N., & Wakslak, C. J. (2026). Relying on AI at work reduces self-efficacy, ownership, and meaning while active collaboration mitigates the effects. Scientific Reports, 16, 13583. https://doi.org/10.1038/s41598-026-42312-6
WRITER. (2026). AI adoption in the enterprise: The invisible crisis in enterprise AI. https://go.writer.com/ai-adoption-enterprise-2026
Duke University Fuqua School of Business. (2025, May 22). Is AI damaging your professional image? https://www.fuqua.duke.edu/duke-fuqua-insights/Is-AI-Damaging-Your-Professional-Image
International Labour Organization. (2025, May 20). Generative AI and jobs: A 2025 update. https://www.ilo.org/publications/generative-ai-and-jobs-2025-update
International Labour Organization. (2026, March 27). New ILO–World Bank paper highlights uneven global impact of generative AI on jobs. https://www.ilo.org/resource/news/new-ilo%E2%80%93world-bank-paper-highlights-uneven-global-impact-generative-ai-jobs
Microsoft. (2025, April 23). The 2025 annual Work Trend Index: The Frontier Firm is born. https://blogs.microsoft.com/blog/2025/04/23/the-2025-annual-work-trend-index-the-frontier-firm-is-born/
Miller, J. (2026, September 29). What do you do with an AI conscientious objector? CIO. https://www.cio.com/article/4227758/what-do-you-do-with-an-ai-conscientious-objector.html
