Should AI Use Be Mandatory at Work?

The short answer

You can require AI-assisted outcomes. You should not require AI usage, and you should not make AI proficiency a promotion or retention gate until you have discharged your training obligation and defined, in writing, what proficiency means in each role.

The distinction matters because the two look identical on a slide and behave in opposite ways in practice. A requirement expressed as an outcome (this analysis is now due in two days rather than five) leaves the method open, is measurable, and survives legal challenge. A requirement expressed as usage (every analyst must log ten prompts a week) measures compliance rather than value, is trivially gamed, and in the European Union now sits close to a category of AI system the law treats as high risk.

What leaders are actually deciding in 2026

This is no longer a hypothetical policy question. It is being decided in performance calibration meetings right now.

In WRITER's 2026 AI adoption in the Enterprise survey, conducted with the research firm Workplace Intelligence across 1,200 C-suite executives and 1,200 non-technical employees who actively use AI at work, 92 percent of the C-suite said they are actively cultivating a class of "AI elite" employees. Sixty percent said they plan to lay off employees who cannot or will not adopt AI. Seventy-seven percent said employees who refuse to become AI-proficient will not be considered for promotion or leadership roles. Ninety percent said the emergence of AI super-users means they will have to completely rethink how they evaluate performance.

The same survey found what the other side of that policy looks like. Twenty-nine percent of employees admitted to sabotaging their company's AI strategy, rising to 44 percent among Gen Z. Seventy-six percent of executives said employee sabotage is a serious threat to the company's future. Only 35 percent of employees said their manager is an AI champion.

Two independent datasets set the context. Pew Research Center found in June 2026 that 71 percent of US adults expect AI to reduce the number of jobs over the next 20 years, up from 64 percent in 2024. And McKinsey's State of AI survey, covering 1,993 respondents across 105 countries and published in November 2025, found that only 39 percent of organisations report any enterprise-level earnings impact from AI, with most of those putting it below 5 percent.

So the pattern is: employees are more worried than they were, most organisations still cannot show the money, and a large share of executives are responding by making individual AI adoption a condition of employment. That is a policy built on pressure rather than evidence.

Why blanket AI mandates backfire

Three mechanisms, each independently documented.

Compliance without capability produces workslop

When people are required to use AI but not equipped to use it well, they produce output that looks finished and is not. BetterUp Labs and the Stanford Social Media Lab named this "workslop" and measured it across 1,150 US full-time desk workers in September 2025: 40 percent had received workslop in the previous month, each incident took around two hours to resolve, and the researchers put the cost at roughly 186 US dollars per employee per month, or about 9 million dollars a year in a 10,000-person organisation.

That cost lands on the recipient, not the sender. A usage mandate rewards the person generating the slop and penalises the colleague who has to fix it, which is precisely the incentive structure you do not want.

Usage metrics become the goal

Any adoption target expressed as tool activity will be met without the underlying work changing. Seats get activated, prompts get logged, and the metric turns green while cycle time, quality and cost stay flat. This is the most common way an AI programme reports success in year one and gets cancelled in year two. If you need the mechanics of measurement that resists this, our guide to managing change during enterprise AI transformation sets out a three-tier approach.

Leaders are mandating against a picture that is already wrong

McKinsey's Superagency in the Workplace research, published in January 2025, found that C-suite leaders estimated only 4 percent of employees use generative AI for at least 30 percent of their daily work, when employees self-reported roughly three times that figure. Looking forward, 20 percent of leaders expected employees to be using AI for more than 30 percent of daily tasks within a year, against 47 percent of employees.

Employees were, on average, ahead of their leadership, not behind it. A mandate aimed at a workforce that is already moving faster than you think does not accelerate adoption. It signals that you have not been paying attention, which is exactly the condition under which quiet non-compliance becomes rational.

What you can defensibly require

The table below separates requirements that hold up from requirements that do not. "Defensible" here means three things at once: it can be justified to an employee, it can be evidenced to a regulator or a tribunal, and it actually moves the business outcome.

The legal boundary, as it stands in August 2026

Two things changed recently and most commentary still has them the wrong way round.

High-risk obligations moved. The literacy duty did not. Under the EU AI Act, Annex III point 4(b) classes as high risk any AI system intended to be used to make decisions affecting the terms of work-related relationships, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate the performance and behaviour of workers. That covers an AI system that scores your employees on AI adoption.

Those obligations were due to apply from 2 August 2026. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It defers the application date for standalone Annex III high-risk systems to 2 December 2027, and for AI embedded in products already covered by EU product safety law to 2 August 2028.

Article 4, the AI literacy duty, was not deferred. It has applied since 2 February 2025 and the Omnibus reworded rather than removed it: providers and deployers must now take measures to support the development of AI literacy among staff, and are explicitly not required to guarantee any specific level of literacy in any individual. That is an obligation of effort rather than result, but it is a live obligation.

Read those two together and the sequencing is set by the law itself. Your duty to help people become AI-literate is in force today. Your ability to run an automated system that judges them on it arrives, with conditions attached, in December 2027. Building the judgement before the capability inverts the order the regulation assumes.

The Singapore dimension. The Workplace Fairness Act, passed on 8 January 2025, with its dispute resolution counterpart passed on 4 November 2025, is expected to commence at the end of 2027, phased in first for employers with 25 or more employees. It makes adverse employment decisions on the basis of protected characteristics unlawful across hiring, performance review, training, promotion and dismissal. Age is one of those characteristics.

AI proficiency is not itself a protected characteristic, and requiring it is not discrimination. The exposure is indirect: Pew Research Center found that workplace AI use skews towards younger workers and those with a bachelor's degree. A promotion policy keyed to AI proficiency can therefore produce an age-correlated outcome without anyone intending it. The mitigation is not to abandon the policy. It is to run the disparate impact analysis before you publish it, and to be able to show that literacy support reached every age band equally.

The sequence: four gates before proficiency counts

This is MASSIVUE practitioner guidance rather than a research finding, drawn from workforce transformation engagements. It is a sequencing rule, and the rule is simple: you may not set a requirement at one gate until the previous gate is demonstrably closed for that population. Populations move through it at different speeds, and that is expected.

Gate 1: Access

Everyone expected to use AI has a sanctioned tool and a clear, written statement of what they may and may not put into it. This gate exists because the alternative is already happening: in the WRITER survey, 35 percent of employees had entered proprietary information into public AI tools and 55 percent described AI use at their company as a chaotic free-for-all. You may require nothing yet. Closing condition: sanctioned tooling and a published data rule for every in-scope role.

Gate 2: Literacy

Role-relevant training delivered, in work hours, with practice time protected in the diary. Generic curricula fail here for reasons we set out in our analysis of why generic AI training fails department by department. This is where the Article 4 duty is discharged. You may require attendance and participation. Closing condition: coverage evidenced across every role and, importantly, every age band.

Gate 3: Outcome

The work standard changes, and it changes in the language of the work rather than the language of the tool. The brief is due in two days. The reconciliation covers the full ledger rather than a sample. The method stays open. You may require the outcome. Closing condition: the new standard is being met by a clear majority without heroics, which tells you the standard is achievable rather than aspirational.

Gate 4: Performance

Only now does AI-assisted capability enter the normal performance conversation, and it enters as one input among many, assessed by a named human decision-maker with a documented appeal route. You may weigh proficiency in progression decisions. If any part of that assessment is automated, you are inside Annex III point 4(b) and the December 2027 conditions apply to you.

What to do instead of building an AI elite

The two-tier workplace is not a strategy. It is what happens when an organisation has no mechanism to spread what its best users already know. Four substitutions do more work than a mandate.

1. Make super-users a distribution channel, not a class. Give the people who are already ahead a formal role in teaching their own function, with time allocated for it. The asset is their method, and the method is worth more shared than hoarded.
2. Fix the manager layer first. When only 35 percent of employees say their manager is an AI champion, a mandate lands on people with nobody to ask. Managers need to be one gate ahead of their teams, not one behind.
3. Answer the time question in public. Employees will not surface efficiency gains they believe will be used to cut their own headcount, and 71 percent of the public already expects AI to reduce jobs. Say explicitly what happens to the time AI frees up, before you ask anyone to free it. The same reticence explains why employees hide their AI use at work even where nothing has been banned.
4. Set a quality bar, not a usage bar. Hold people accountable for what they send, whether or not AI helped produce it. This is the single most effective control against workslop, and it requires no policy change at all.

Frequently asked questions

Can we dismiss an employee who refuses to use AI?

Treat it as a performance or role-fit question, never as an AI-adoption disciplinary category. If the role's outcome standard has legitimately changed, has been communicated, and the employee has been given the tool, the training and reasonable time, then persistent failure to meet the standard is an ordinary performance matter under your local employment law. Framing the same decision as "refused to adopt AI" converts a defensible performance case into a novel and untested one, and in the EU it points at an AI system category the law treats as high risk. Take local legal advice before acting.

Should AI proficiency appear in performance reviews?

Eventually, yes, but only after literacy has been delivered and proficiency has been defined for that specific role in writing. Assessed earlier, you are measuring who had access to good training and who had a manager who understood the tools, which is a management failure being recorded as an employee failure.

Does the EU AI Act stop us from using AI to evaluate employees?

No, it conditions it. Annex III point 4(b) classifies such systems as high risk, which brings obligations on risk management, data governance, human oversight, transparency and record keeping rather than a prohibition. Following Regulation (EU) 2026/1744 those obligations apply from 2 December 2027 for standalone systems. The AI literacy duty in Article 4 applies now.

Can an employer require the use of a specific AI tool?

Narrowly, yes. Requiring a named sanctioned tool is defensible for a specific workflow where audit trail, data residency or confidentiality demands it, and the reason should be documented for that workflow. Extending a tool mandate to general work measures compliance rather than value and is gamed quickly.

Where MASSIVUE fits

MASSIVUE works with enterprises on AI workforce transformation: skills assessment and gap analysis, role design, and the capability programmes that close the gates above in the right order.

If you are the person who has to write this policy, the closest fit in MASSIVUE Academy is the Applied AI HR Talent Management micro-credential, which covers AI across hiring, promotion, pay and performance together with the EU high-risk obligations that follow. For HR leaders who want the full remit rather than a single module, the Certified AI HR Specialist certification covers bringing AI into hiring, development and workforce decisions without losing fairness, privacy or human judgement. If your immediate problem is the resistance rather than the policy, AI Change Management: Upskilling & Reskilling is the more direct route.

Sources

• WRITER and Workplace Intelligence, 2026 AI adoption in the Enterprise survey, 7 April 2026. 1,200 C-suite executives and 1,200 non-technical employees actively using AI at work. Vendor-commissioned. https://writer.com/blog/enterprise-ai-adoption-2026/
• Kate Niederhoffer, Gabriella Rosen Kellerman and colleagues, AI-Generated "Workslop" Is Destroying Productivity, Harvard Business Review, 22 September 2025. Research by BetterUp Labs with the Stanford Social Media Lab, 1,150 US full-time desk workers, September 2025. https://hbr.org/2025/09/ai-generated-workslop-is-destroying-productivity
• BetterUp Labs, Workslop: The Hidden Cost of AI-Generated Busywork. Methodology and cost figures. https://www.betterup.com/workslop
• McKinsey & Company, Superagency in the workplace: Empowering people to unlock AI's full potential at work, January 2025. Leader and employee perception gap. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work
• McKinsey & Company, The State of AI: Global Survey 2025, 5 November 2025. 1,993 respondents across 105 countries. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
• Pew Research Center, Young adults in the U.S. are increasingly wary of AI, concerned it will take jobs, 18 August 2026. 3,488 US adults, fieldwork 22 to 28 June 2026. https://www.pewresearch.org/short-reads/2026/08/18/young-adults-in-the-us-are-increasingly-wary-of-ai-concerned-it-will-take-jobs/
• Pew Research Center, U.S. Workers Are More Worried Than Hopeful About Future AI Use in the Workplace, 25 February 2025. 5,273 employed US adults, fieldwork 7 to 13 October 2024. Source for AI use skewing towards younger and degree-holding workers. https://www.pewresearch.org/social-trends/2025/02/25/u-s-workers-are-more-worried-than-hopeful-about-future-ai-use-in-the-workplace/
• Regulation (EU) 2024/1689 (the AI Act), Annex III point 4 and Article 4. https://artificialintelligenceact.eu/annex/3/
• Regulation (EU) 2026/1744 (the Digital Omnibus on AI), published in the Official Journal 24 July 2026, in force 27 July 2026. Deferral of Annex III high-risk application to 2 December 2027 and Annex I to 2 August 2028; amendment of Article 4. https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/
• Singapore Workplace Fairness Act, passed 8 January 2025, and the Workplace Fairness (Dispute Resolution) Act, passed 4 November 2025. Commencement expected end-2027, phased from employers with 25 or more employees. https://www.tal.sg/tafep/workplace-fairness

Legal content is general information current at 19 August 2026 and is not legal advice. Obligations differ by jurisdiction and the EU timetable has already moved once.