AI and workplace safety: risks, benefits and what to do

AI is entering workplaces not only as a writing or analysis tool but as a manager: software that assigns tasks, sets the pace, tracks performance and schedules shifts, and robots that work alongside people. That changes health and safety at work. This page summarizes what the European Agency for Safety and Health at Work (EU-OSHA) has found and what employers and workers can do about it.

The short answer: AI can make work safer, for example by spotting hazards, flagging poor posture or helping plan training. It can also make work more stressful and riskier when it pushes the pace, removes breaks, reduces control over the job and puts people under constant monitoring. EU-OSHA's research points to the same remedy again and again: involve workers in how these systems are chosen, set up and checked, and keep people in command.

What EU-OSHA means by AI worker management

EU-OSHA uses the term AI-based worker management for systems that gather data, often in real time, on the workplace, the workers, their work and their tools, and feed it into an AI model that makes automated or semi-automated decisions or gives information to managers (EU-OSHA, AI for worker management, 2022). Examples include systems that allocate tasks, set routes or schedules, and score performance.

Where AI can help safety

The same report lists opportunities: better monitoring of hazards, a system that directs work while also alerting workers to poor posture and the risk of musculoskeletal disorders, adjusting tasks or schedules for workers with health needs, and support for designing safety training and safety strategies (EU-OSHA, 2022).

The risks EU-OSHA identifies

• Pace and intensity: systems that direct workers can push them to skip breaks or work at high speed.
• Loss of control: less autonomy over how work is done, linked to high stress.
• Performance pressure: systems that monitor and evaluate workers, with risks of musculoskeletal disorders, exhaustion, accidents, stress and anxiety.
• Dehumanisation: strict control that treats people as data points and asks them to work like machines.
• Privacy and work-life balance: intrusive monitoring that collects private data and blurs the line between work and home.
• Discrimination: systems built on biased data that favour workers of a particular age, ethnicity or gender.

These points come from the conclusions of EU-OSHA's report (EU-OSHA, AI for worker management, 2022). A later EU-OSHA paper names work intensification and stress from constant surveillance and data gathering as the main threats (EU-OSHA, Worker participation and AI worker management, 2024).

Robots and cobots

EU-OSHA's review of advanced robotics found them used for tasks such as lifting or assisting patients in nursing, and for welding, assembly, packaging, transport and delivery in manufacturing, hospitals and warehouses. How tasks are split between people and robots matters: done well, it can reduce errors and balance workload; done badly, it can lead to over-reliance, mistrust, reduced situation awareness and an unbalanced mental workload (EU-OSHA, Advanced robotics and automation, 2022). See will AI replace warehouse workers? and will AI replace factory workers?

What employers can do

EU-OSHA recommends a prevention-through-design, human-centred approach (EU-OSHA, 2022):

• Keep humans in command: use AI to support workers, not replace their judgment, and make the system transparent and understandable.
• Consult and involve workers at every stage, from choosing a system to evaluating it.
• Adapt the risk assessment: cover all work-related factors, bring in people who understand the algorithm, and repeat the assessment, because systems that learn can change.
• Train managers and workers on how the system works and how it can affect health and safety, and give workers a way to get support.
• Collect only what is needed and comply with existing law, including safety, data protection and anti-discrimination rules.

In the EU, AI used to allocate tasks based on behaviour or personal traits, or to monitor and evaluate workers, is listed as high-risk (EU AI Act, Annex III, point 4(b)), and employers will have to inform workers' representatives and affected workers before using such a system (EU AI Act, Article 26(7)). These high-risk duties apply from 2 December 2027, the date set by the Digital Omnibus on AI, Regulation (EU) 2026/1744 (European Commission; NI Assembly EU law tracker). Safety, data protection and anti-discrimination law already apply.

Why worker involvement matters

EU-OSHA's 2024 case studies found that involving workers' representatives helped limit risks: at a Swedish mining company it addressed the use of personal data for performance and working-hours monitoring, and at a Danish manufacturer it helped refine smartwatches used for task allocation. It also found that representatives struggle when the technology is opaque, and recommends giving them real access to the algorithms and data (EU-OSHA, 2024). Some collective agreements now list safety around AI-controlled equipment as a topic for regular union meetings, for example the 2024 IATSE Basic Agreement for film and TV crews (2024 IATSE Basic Agreement, Article L); see AI and unions.

What workers can ask

• What does the system decide, and what does a person decide?
• What data does it collect about me, and who sees it?
• Does it set my pace or my breaks, and can I raise a concern if it is unsafe?
• Was the safety risk assessment updated when it was introduced? For the legal duties in Great Britain, see health and safety duties when algorithms manage work.

See also AI monitoring at work, AI for managers and will AI replace safety specialists? This page is general information, not legal or safety advice.