The blind spot on AI at work in Europe

Europe aims to integrate AI in factories and workplaces. The question of governance arises, states Aida Ponce De Castillo, Senior Researcher at the European Trade Union Institute

Two accounts of artificial intelligence (AI) ran side by side in Ursula von der Leyen’s State of the Union speech delivered at the beginning of Autumn 2026, in Strasbourg: one on frontier AI, the other – quieter – on its arrival in everyday work. Most workers will encounter the second.

One concerned frontier AI. She warned about increasingly capable models, cyber risks, self-improving systems and AI agents behaving in unexpected ways. Europe, she argued, needs both safeguards and enough technological capacity to remain in the race.

The other was much closer to everyday economic life. Von der Leyen wants Europe to move AI “from the screen to the real economy and society”, including “on the factory floor”. She listed hospitals, agriculture, energy, autonomous driving, defence, and advanced manufacturing. She recognised that Europe does not have to produce a leading frontier model to capture economic value from AI.

Much of that value may come from applying it. This is an important change of emphasis. Tech Policy Press picked up the same point, describing the speech as a move towards an AI strategy where deployment, industrial capacity and technological sovereignty matter at least as much as leadership in frontier models. If European policy is heading that way, the workplace becomes much more important to the AI debate.

Beyond the jobs question

Most public discussion about AI and employment still gravitates towards a familiar question: how many jobs will disappear? It is indeed an important question, but it captures only one part of what is happening. A job may remain formally, but its conditions, pace, autonomy and modes of evaluation could change substantially.

Software can allocate tasks, establish priorities, monitor activity, recommend schedules or feed information into performance assessments. A manager may retain formal authority while relying increasingly on a system to tell her who is performing well or to flag workers for intervention. In short, AI can reshape how work is done.

Some of these practices fall under what is commonly called algorithmic management, although the term should be used carefully. Algorithmic management is broader than AI. A relatively simple digital monitoring or scheduling system may exercise a managerial function without containing anything ordinarily described as artificial intelligence. At the same time, the growing integration of AI systems into management functions makes it important to distinguish between AI-based management systems and other forms of algorithmic management, since they may fall under different regulatory regimes.

This is not a marginal phenomenon. A 2024–2025 survey by the European Commission’s Joint Research Centre, covering more than 70,000 workers across the EU, found that almost one in four workers already have their working time allocated automatically, while 13% report that their performance is assessed and rewarded automatically. A 2025 European Parliamentary Research Service analysis of Eurofound data estimates that 42.3% of EU workers may be subject to some form of algorithmic management.

Often, what matters for workers is less the label attached to the technology than what the system really does. For example, a system that helps a worker find information is one thing. A system that measures her output, compares her with colleagues, recommends a performance rating or determines the sequence in which she must perform tasks – whether on a production line, in a warehouse, on a university campus, in an automated harbour, or working alongside humanoid robots on an industrial site – occupies a very different position in the employment relationship.

The productivity trap

This is where von der Leyen’s insistence on “AI with human agency” becomes interesting. Human agency sounds reassuring, but its meaning at work cannot be taken for granted. The presence of a human manager somewhere in a decision-making process does not necessarily result in meaningful human control from a labour perspective. If a manager routinely accepts a system’s recommendation, either because the system is opaque, because there is no time to examine it or because organisational practice discourages deviation, human intervention can become little more than a procedure.

Human agency also has more than one possible subject. There is the agency of the manager using the technology, but there is also that of the worker subjected to it. And where technologies reorganise work collectively, a further issue concerns workers’ and their representatives’ capacity to understand, discuss, and influence the systems before they are introduced.

These questions matter when AI is promoted as a route to higher productivity. Productivity is not a neutral variable inside a workplace. Someone decides what is measured, over what period and according to which benchmark. The choice of indicators can influence behaviour long before an automated system makes any formal decision. A technology designed to optimise output can reshape work even when it never hires, fires or disciplines anybody.

From an occupational safety and health (OSH) perspective, productivity metrics can themselves become occupational risk factors when they determine work pace, recovery time, movement or the degree of discretion workers retain over how safely a task can be performed.

There are potential gains. AI can remove repetitive tasks, reduce physical workload, assist workers with complex information, identify hazards in the working environment and make some processes more efficient. There is no reason to assume that workplace AI must necessarily make jobs worse.

There is equally no reason to assume that productivity gains will automatically improve job quality. The same technologies can increase work intensity, extend monitoring, narrow worker autonomy, or make forms of managerial control less visible. The outcome depends on design choices, organisational practices, bargaining power and regulation. The 2025 ILO report on AI, digitalisation and occupational safety and health captures this dual effect. These technologies can help reduce hazardous exposures and injuries, but they can also generate new risks.

AI-enabled systems can affect task allocation, work pace, autonomy and workers’ exposure to physical and ergonomic risks. Interaction with robotics and other automated systems can introduce additional hazards, including ergonomic problems, system failures and unexpected machine behaviour. Growing reliance on automated systems raises another OSH question: whether workers retain sufficient agency, oversight, and room to intervene when something goes wrong. This is why emerging risks require co-constructed preventive responses.

Workers and their representatives need a clear role in deciding what problem the technology is supposed to solve, which tasks should be automated or supported, which indicators should be used, what level of work intensity is acceptable and where human discretion must be preserved.

A preventive approach also means testing systems against the reality of the job, incorporating occupational risks into OSH, AI and data-protection assessments before deployment, and retaining the possibility of modifying or withdrawing a system when its effects on work become apparent. The ILO explicitly places worker participation and preventive strategies among the elements needed to safely integrate digital technologies at work.

This practical understanding of human agency also connects back to the political ambition set out by von der Leyen. After setting out her ambitions for industrial AI, she turned to its effects on Europe’s social market economy and said AI should “improve the quality of jobs”. If that objective is taken seriously, worker agency has to extend to the choices that determine how the technology will organise, measure and govern work.

What the law already covers – and doesn’t

EU law already reaches some of these practices. The GDPR applies when workers’ personal data are processed and provides safeguards relevant to automated decision-making. Under the AI Act, several employment-related uses of AI, including certain systems used for recruitment, task allocation, monitoring, and performance evaluation, are classified as high-risk. Data protection, anti-discrimination, product-safety and other rules are one layer; how work itself is organised is another, and that is where labour law comes in. Labour law adds rights of information and consultation in circumstances where technological change affects the organisation of work. This is not a regulatory vacuum. But neither is it a completely coherent framework for the growing use of digital systems as instruments of management.

The AI Act regulates AI systems according to its own architecture and definitions, which stem from a product-safety perspective. It does not cover every form of algorithmic management. Data protection law also has a different object and logic.

The Commission appears to recognise the problem. In July, it opened the second stage of consultation with European trade unions and employers on its forthcoming Quality Jobs Act. The first area listed for possible EU action is “algorithmic management and artificial intelligence at work”, including greater transparency around automated decisions and protection against excessive monitoring. That initiative should be watched, although much will depend on what eventually appears in the legislative proposal.

What is clear is that Europe wants AI to spread much further into the economy while still describing its model as human-centred. That claim will be tested in ordinary workplaces, where decisions about automation, monitoring and control are made every day.

If von der Leyen’s industrial strategy succeeds, many workers will encounter AI through systems that organise schedules, measure productivity, assess performance or monitor activity. These uses are far less spectacular than frontier models or humanoid robots, but they may have a much more immediate effect on how work is organised and experienced. This is also where the economic value promised by AI will be produced. It is, therefore, where questions of autonomy, responsibility and managerial power become unavoidable.

Europe wants more AI on the factory floor. The issue is whether the rules governing its use will develop at the same pace as the technology itself.