An AI tool does more than produce an output. It can change who makes a decision, what information they see, the time available to respond and who is accountable when a recommendation is wrong.
Assess the changed task, not only the model
EU-OSHA identifies both opportunities and risks from digitalisation, including issues in human-machine interaction, overreliance, worker monitoring and AI-based management. The direction of impact depends on how a system is implemented. [1]
Map the task before and after deployment. Which decisions disappear, which move to a worker and which become harder to question? Consider a hypothetical alert system that reduces routine checking but adds a new responsibility to distinguish urgent signals from false alarms.
Make oversight operational
"Human in the loop" is not a complete control specification. Identify the person who can intervene, the evidence they receive, their authority to disagree and the time needed to act. Test what happens when the system is unavailable or produces a plausible but incorrect answer.
A professional should not have to choose between challenging an output and meeting an impossible response target. The surrounding work design belongs in the assessment.
Listen before collecting more data
Explain the purpose of monitoring, what information is collected, who can access it and how it can affect a worker. Do not assume that a safety label removes privacy or employment concerns. EU-OSHA highlights consultation and transparency in its discussion of AI-based worker management. [1]
Ask workers to describe how a new system changes pace, autonomy, support and the ability to report a problem. Record intended benefits and potential adverse effects separately.
Review actual effects
Select indicators connected to the claimed benefit. A higher number of alerts is not, by itself, evidence of improved safety. Pair operational data with observation, worker feedback and a process for escalation.
This is an assessment prompt, not a legal determination or certification scheme. Applicable AI, privacy, employment and safety requirements need a separate jurisdiction-specific review.
Sources & context
- EU-OSHA: Digitalisation of work EU-OSHA · European Union
- HSE: Human factors and ergonomics Health and Safety Executive · Great Britain
Source-linked editorial material, prepared 2026-09-05; updated 2026-09-05. Prepared with AI assistance. Not independently peer-reviewed. Sources and legal requirements can change. The linked organisations do not endorse this publication. Our editorial approach.
What would you add?
Where has automation helped your team, and where has it created a new kind of work?
Connecting to the discussion...