AI is a human challenge, not an IT challenge.
British Chambers of Commerce’s Future of Work: AI in the Workplace Report found that 54% of UK firms now use it, up from 23% in 2023. The Department for Science, Innovation and Technology (DSIT)’s AI adoption survey report 2026 discovered most are adopting it for efficiency and productivity. Yet only 25% of employees say their organisation has communicated a clear plan for AI.
Generative AI tools are arriving into a workforce that is already stretched. Average sickness absence has reached 9.4 days per employee per year, and heavy workloads remain the top cause of stress-related absence, according to CIPD’s Health and wellbeing at work report.
Over time, most organisations will have access to the same AI tools. What will set them apart is how well their people use them, and that depends far more on how AI is introduced than on the tool itself.
Below are five areas for HR and managers to consider in terms of AI workplace safety and management. Important definitions are explained in our AI Glossary for HR and Managers.
1. Introduce governance
Effective AI management in the workplace starts with a clear standard rather than a ban. 71% of UK employees have used “shadow AI” (unapproved consumer AI tools at work), and 51% still do every week. 28% say it is because their employer doesn’t offer an approved option. This is most often because people are trying to keep up without clear guidance. Banning tools tends to push use underground, whereas a clear standard brings it into the open.
A good AI standard covers:
- Approved tools and where AI can be used, including what information can and cannot be shared (data handling and prompt guidance)
- Where AI cannot be used, for example highly sensitive data, formal employee relations documentation or regulated decisions
- Human oversight: who checks AI outputs, and how they are documented and challenged
- Automated decisions: in line with the Data (Use and Access) Act 2025UK employers making solely automated decisions with significant effects on people, such as in recruitment or dismissal, must tell individuals, offer meaningful human intervention and allow them to contest the decision
- Consequences of unauthorised or inappropriate use, aligned to your disciplinary policy
Keep the headline simple enough to fit on one line, for example: “approved tools”, “no personal data”, and “human signs off”.
2. Address the psychological effects
Have you considered the psychological impact of AI, AI’s effect on mental health, or the risk of burnout? There are three distinct forms of AI anxiety, and each needs a different response.
- Existential anxiety (AI job insecurity): “What will AI do to my role and my profession over time?” This can’t be fixed with reassurance. Be honest about what you do and don’t know yet. False reassurance makes it worse.
- Operational anxiety (a form of technostress): the pressure of using new tools now, at a pace people didn’t choose. This is the most fixable. Give people time to learn, permission to be a beginner and a say in the pace.
- Identity anxiety (professional identity anxiety): “If AI can do what I do, who am I professionally?” Your most experienced people are likely to be the most reluctant adopters. Be clear about how they bring value, and move them from producing work to judging it.
Applying the Job Demands–Resources (JD–R) model to AI is a useful way to think about this. AI adds new demands: learning new tools, checking outputs and, in some workplaces, being monitored by AI. It can also add resources: less admin, more autonomy and better information. Whether AI improves or harms wellbeing depends largely on whether the resources keep pace with the demands.
AI workplace safety also extends to monitoring employees’ needs with particular care. For example, Co-op Legal Services faced criticism over its use of AI to monitor staff calls with customers. Whatever the intent, being constantly assessed by AI can feel like surveillance, and that is a recognised psychosocial risk. Be explicit about what is and isn’t monitored, and why.
All of this rests on psychological safety: people need to feel able to ask for support and raise concerns without fear.
3. Build AI confidence, productivity and literacy
Useful, reliable tools
When employees see the AI tools they use as useful and reliable, it is linked to higher productivity, engagement and job satisfaction. A 2025 study of 600 employees, using the JD–R model, found AI efficacy worked as a resource. AI technostress worked as a demand, linked to exhaustion, work-family conflict and lower job satisfaction. Choosing the right tools, and being honest about where they don’t work well, matters as much as rolling them out.
People also need to know when a tool is reliable enough to trust for a task, when to check its output and when to override it.
AI self-efficacy
Confidence matters too. AI self-efficacy is a person’s belief in their ability to learn and use AI effectively, which is different from how well the tool itself performs. Confidence grows through hands-on use in a safe setting, not a one-off training session.
AI productivity
Rather than measuring time saved, measure what people have done with that time.
Let’s take an example. Two employees each use AI to save five hours a week. One is told to fill those hours with more work, whereas the other is encouraged to use them for higher-value work, learning or recovery. The first is heading for burnout, while the second will likely experience better engagement. Decide up front where saved time goes.
AI literacy
Employees need the skills to use AI well, including critical thinking and knowing what information they can share. Managers need more than that. They are being asked to lead AI adoption and support wellbeing at the same time, often trained for neither. In fact, only 45% of managers are trained for mental health conversations .
A practical way to start is by removing low-value work: ask each person to pick one task they resent that is low-stakes and regular, and try AI on it. The choice should be theirs. That protects time for the things that need people: judgement, coaching and relationships.
4. Review roles, responsibilities and workflows
AI changes tasks before it changes jobs. Break roles down into their individual tasks and ask which could be automated, which AI could support and which need to stay with people. Then redesign the role around the result, rather than adding AI on top of an already full workload.
Be clear about accountability, as using AI doesn’t move responsibility for the work. A simple standard for managers to set: “I don’t mind how it got made. I mind that it’s good, that you’ve checked it, and that you’ll stand behind it.”
Watch out for work creep, where roles quietly expand because AI makes more possible. Always-on can’t be solved by culture messaging alone. Build decisions into AI policy. Working hours, response expectations and where saved time goes.
AI could even reshape the HR function itself. Perhaps Human Resources will become Human & AI Resources (H&AIR). A new email signature, perhaps?
5. Involve employees and support organisational change
A common mistake is treating AI adoption primarily as a technology project rather than a people-related change. People cope far better with change they have helped shape, so make sure to involve them before you implement any big changes.
As part of AI workforce safety and management best practice, make sure to:
- Ask: find out what people are using AI for, what has worked and what worries them, then close the loop on what you heard
- Create AI champions: a network of colleagues who share what works and help others get started
- Equip line managers: for both conversations, the AI one and the wellbeing one
- Lead by example: leaders who use AI openly, and share what didn’t work as well as what did, give others permission to experiment
AI workplace safety risks for HR
Employee relations
Employees increasingly use AI to draft grievances, subject access requests and tribunal claims, which can be lengthy and complex to respond to. It’s worth understanding the unintended consequences of AI in employment and how AI is reshaping workplace grievances.
Discrimination and data protection
Bias in AI tools used for recruitment or performance decisions can amount to discrimination under the Equality Act 2010, and the employer remains responsible even when the tool comes from a supplier. Personal and sensitive employee data should never go into unapproved tools.
Wellbeing
Employees are turning to general-purpose AI chatbots for emotional support. These tools aren’t designed or clinically validated for mental health support, and there can be negative effects of AI chatbots on mental health. Make sure people know where to find help from a human, such as your Employee Assistance Programme, mental health first aiders or their GP.
AI use cases for HR
Alongside managing AI workplace safety, HR teams can also use AI tools to improve their own processes.
- Analyse employee survey data to spot themes faster, using approved tools and anonymised data
- Set up a regular AI summary, for example a weekly roundup of employment law updates
- Create an AI assistant that answers employee questions on policies, benefits and wellbeing initiatives, built on your own documents (a grounded knowledge base) and checked regularly for accuracy
Five things to do for successful AI workplace safety and management
- Get people’s insights: what are people using AI for, what would they never do again, and what worries them?
- Agree a one-line AI standard, and the few boundaries that sit around it.
- Brief your line managers on both the AI conversation and the wellbeing conversation.
- Pick one task people resent: low-stakes and regular, chosen by the person who does it.
- Use AI yourself, openly, and share what you learn, failures included.
AI workplace safety and management is all about how you adopt the technology. Early decisions can have a big impact on whether AI helps or harms your people, which is why it’s so important that HR are involved and not just IT teams. Arguably, if HR can take the lead with AI adoption businesses may very well see much more successful human adoption.
Are you looking to implement AI safely and effectively?
Altruist are currently researching the people impact of AI at work, focusing on how organisations are managing the impact on employees rather than the technical implementation. We’re inviting HR professionals and business leaders to share their experiences in a 30-minute conversation. If you’d like to take part, please email info@altruistuk.com. As a thank you, we’ll share an anonymised summary of the findings.
