Employment Law News Roundup – 5.9.26

Workplace AI and the Risk of Managerial De-Skilling

New research led by academics at the University of Bath has warned that over-reliance on workplace AI could weaken the practical judgement managers develop through experience. The concern is not that managers should stop using generative AI, but that repeatedly outsourcing thinking to it may reduce the human skills needed to deal with complex workplace situations.

The research focused on “managerial phronesis” - practical wisdom built through experience, reflection and interaction with other people. They argue that workplace AI cannot acquire that experience itself: it can generate plausible answers from existing information, but it does not experience the consequences of decisions or understand workplace relationships in the way a manager does.

When AI Starts to Replace Judgement

The principal risk identified is “epistemic de-skilling”. This can occur when managers use workplace AI as a shortcut for problem-solving and idea generation, particularly when they are under time pressure. If managers increasingly accept AI-generated answers rather than testing them against experience, seeking other perspectives or working through difficult issues themselves, their ability to develop judgement may weaken over time.

That concern is supported by other research. Microsoft researchers surveyed 319 knowledge workers and examined 936 examples of generative AI use at work. They found that greater confidence in AI was associated with less critical-thinking effort. The nature of critical thinking also shifted: workers spent more effort checking AI outputs, integrating them into their work and supervising the task, rather than carrying out the original thinking themselves.

This does not mean workplace AI is inherently detrimental. Separate field research involving more than 7,000 knowledge workers found genuine productivity benefits, including less time spent on email among employees who used generative AI. The issue for employers is therefore not whether AI can save time, but whether efficiency is achieved by assisting human judgement or replacing too much of the thinking that develops it.

The Bath researchers identify a more positive outcome where AI is used to challenge rather than replace judgement. Managers can use workplace AI to test assumptions, generate alternative scenarios and expose gaps in their own reasoning. Accountability appears to be important: where managers know they must explain and justify a decision, AI is more likely to prompt reflection rather than become a substitute for it.

Why This Matters for Employers

The distinction becomes particularly important when managerial decisions affect employees. Recruitment, promotion, performance management, disciplinary action, grievances and redundancy exercises can involve competing evidence, individual circumstances and questions of fairness. An apparently polished AI recommendation may overlook context that an experienced manager would recognise.

The regulatory direction is also towards meaningful human oversight. The Information Commissioner's Office has recently examined automated decision-making in recruitment and identified the need for greater transparency, consistent human involvement and better monitoring for unfair or biased outcomes. Although that work concerns automated recruitment rather than ordinary generative AI use by managers, the underlying principle is relevant: workplace AI should support accountable decision-making, not obscure who actually made the decision.

For employers, the practical response should be to distinguish between tasks that can safely be accelerated and decisions that require human judgement. Workplace AI may be well suited to summarising information, producing initial drafts, identifying questions or generating options. Greater caution is appropriate where an output could materially affect an employee or where the manager must assess credibility, context, fairness or proportionality.

Managers should also be expected to verify important outputs and be able to explain the reasoning behind their decisions without simply referring back to an AI-generated answer. Training should therefore go beyond how to prompt an AI tool. It should include how to challenge outputs, identify missing context, check factual accuracy and recognise when direct human engagement is required.

The implication is not that employers should resist workplace AI. Used well, it can improve efficiency and widen the range of information considered. The risk arises when efficiency becomes cognitive outsourcing. Organisations that preserve human accountability and deliberately require managers to exercise judgement are more likely to gain the benefits of AI without weakening the skills on which good management depends.

Businesses Confront Risks of Unauthorised Workplace AI

Businesses are increasingly having to confront the use of workplace AI without organisational approval or oversight. Research commissioned by Turbotic found that 88% of senior decision-makers surveyed believed unauthorised AI tools were being used within their organisation, while 51% were concerned that sensitive company information was being entered into those tools.

Those figures reflect the perceptions of senior decision-makers rather than independently verified employee behaviour, but the underlying problem is supported by other evidence. Microsoft reported in 2025 that 71% of UK employees had used unapproved consumer AI tools at work and 51% were doing so every week.

Shadow AI and Workplace AI Governance

Unauthorised use is commonly described as “Shadow AI”. It can arise when employees use public chatbots or other workplace AI tools outside the systems approved by their employer.

The Turbotic survey suggests the practice is not confined to one part of the business. Respondents reported unauthorised AI use most frequently in IT at 36%, followed by customer service at 30%, marketing at 28% and sales at 27%. Their principal concerns included regulatory or compliance breaches, lack of management oversight and unreliable AI-generated information influencing business decisions.

Official UK data also points to a wider governance gap. The UK Business Data Survey 2026 found that 41% of businesses handling digitised data were using AI for at least one purpose, but only 17% of AI-using businesses reported having either a formal policy or informal guidance governing its use. Workplace AI is therefore becoming established considerably faster than formal governance in many organisations.

That matters because employees can adopt AI without procurement, IT deployment or a formal business project. A publicly available chatbot can be used immediately to summarise a document, draft an email, analyse information or prepare a report. The employer may have no visibility over which tool is being used, what information has been entered or what happens to that information afterwards.

Why Employees Use Unauthorised AI

Treating Shadow AI simply as employee misconduct risks missing an important part of the problem. National Cyber Security Centre guidance on shadow IT recognises that unauthorised technology is rarely adopted with malicious intent. Employees often turn to unofficial tools because approved systems are slow, difficult to access or do not provide the functionality needed to complete the task.

The NCSC specifically identifies the absence of approved AI functionality for tasks such as rewriting documents, compiling information or summarising meetings as one reason employees may turn to shadow systems.

That has an important implication for workplace AI policies. A blanket prohibition may look clear on paper but prove ineffective if employees can obtain a significant productivity advantage from tools that the organisation has not provided. Employers need both controls and workable approved alternatives.

Data Protection, Confidentiality and Accuracy

The most immediate risk is often information leaving the organisation. Employees may paste customer information, employee data, contracts, internal correspondence or commercially sensitive material into an external AI service without understanding how the provider processes or retains it.

Where personal information is involved, existing data protection obligations continue to apply. ICO guidance makes clear that organisations using AI to process personal data must comply with UK data protection law, including requirements concerning fairness, transparency, security, data minimisation and accountability.

Confidentiality and intellectual property can create separate concerns even where personal data is not involved. Employers also face the risk that inaccurate AI-generated material is incorporated into correspondence, reports or decisions without adequate checking. The National Cyber Security Centre continues to warn that generative AI can produce incorrect information confidently and can introduce additional security risks.

The problem can become more serious where unauthorised workplace AI influences decisions about employees. An organisation may believe a decision was made by a manager when, in practice, an unapproved AI tool materially shaped the analysis. That creates obvious difficulties around transparency, consistency and accountability.

What Employers Should Do About Shadow AI

The starting point is visibility. Employers need to understand which workplace AI tools are actually being used, by whom and for what purposes. That is likely to require engagement with employees as well as technical controls; staff are less likely to disclose informal use if the exercise is presented solely as an investigation into wrongdoing.

Policies should then distinguish clearly between approved and prohibited uses. Employees should know which tools they may use, what categories of information must not be entered into public systems and which decisions require human review. Rules should be practical enough to apply to everyday tasks rather than relying on broad statements that “confidential information” must not be shared.

Employers should also review the tools they provide. If an approved workplace AI service can perform the tasks employees are already trying to accomplish, there is less incentive to bypass organisational controls. Procurement should consider data handling, security, contractual terms and whether information submitted to the service may be retained or used for other purposes.

Training is equally important. Employees need to understand not only what the rules are but why they exist: AI outputs can be wrong, confidential information can leave the organisation and personal data cannot simply be uploaded because doing so is convenient.

The objective should not be to eliminate workplace AI. In many businesses that is neither realistic nor desirable. The more effective approach is to bring actual use into the organisation's governance framework so that employees have useful approved tools, clear boundaries and appropriate human oversight.