Artificial intelligence was once sold as a tool to make work easier. New research from TU Darmstadt suggests that those who know it best are far less optimistic. A representative survey of more than 2,000 people in Germany, conducted with YouGov, shows that the strongest AI knowledge goes hand in hand with the greatest fear of job loss.
Understanding AI increases fear of job loss
The survey found that 43% of respondents with very strong AI knowledge expect artificial intelligence to take over their jobs in the foreseeable future. That figure is considerably higher than among people with moderate or little understanding of the technology. In other words, familiarity with AI does not breed comfort. It breeds awareness of how many tasks can be automated.
Peter Buxmann, who led the study at TU Darmstadt, described the pattern as remarkable and worrying. He noted that people who understand AI best are also the most aware of its disruptive potential. The more one knows about current capabilities, the less easy it is to dismiss the technology as overhyped. Large language models can draft contracts, generate code, summarize documents, and analyze datasets in seconds. It is no longer science fiction to imagine a workplace where many of these jobs are done by machines.
The survey was designed to be representative of the German population. Its findings were compared with last year's edition, which allowed the researchers to measure how attitudes have shifted over time. The comparison revealed that concern about job security rose across every occupational group since the previous survey. This suggests that the anxiety is not limited to one industry or one type of worker. It is spreading.
Output matters more than industry
One of the more interesting findings is that the sector in which someone works matters less than what they actually produce. Workers who generate text, analysis, reports, code, or presentation slides appear to be more exposed to AI-driven disruption than those who perform physical or interpersonal tasks. The reason is simple: these are the kinds of outputs that current AI systems handle with increasing competence.
A data analyst who writes reports may find that AI can perform the core task in seconds. A customer service representative who answers repetitive questions may face similar pressure. But a nurse, electrician, or childcare worker has little to fear from a chatbot. This distinction helps explain why AI anxiety is not evenly distributed across the labor market. It also helps workers understand whether they should be worried at all.
The fears are practical, not existential
Another surprise in the survey concerns the nature of the worries themselves. The most common fears have nothing to do with machines becoming superintelligent or taking over the world. Instead, respondents cited hallucinations and data protection as their top concerns. They worry that AI systems will make things up, and they worry about what happens to the data they feed into these systems.
This is a very practical set of concerns. Hallucinations are a known problem with large language models. A model may generate a confident answer that contains false dates, invented quotes, or made-up legal references. If an employee uses such an output in a report or an email, the consequences can be embarrassing at best and legally dangerous at worst. Data protection is equally important, especially in Germany, where privacy laws are strict and public attitudes toward data collection are cautious.
Only 40% of respondents considered superintelligence a likely outcome. The majority are less concerned with some future artificial general intelligence and more focused on the way AI behaves today. This is a useful corrective to the science-fiction framing that dominates much of the media. For most workers, the risk is not that an AI will wake up and decide to eliminate humanity. The risk is that an AI will replace them in their daily tasks long before anything resembling consciousness arrives.
Younger workers are more anxious
The survey also inverts the usual generational expectation. Younger workers are more anxious about job security than older workers. Traditional assumptions would suggest that older employees, who may be less familiar with digital tools, would be the most fearful. But the data shows the opposite. Younger workers have grown up with technology, yet they seem more aware of what it can do to their careers.
There are several possible explanations. Younger workers are more likely to be early in their careers, with less job security and more to lose. They are also more likely to be employed in white-collar roles that are directly exposed to AI automation. In addition, they may have a more realistic understanding of AI's capabilities because they use it regularly. The result is that age does not protect against anxiety. If anything, youth amplifies it.
The training gap is the biggest problem
While the fear of job loss may be understandable, the survey uncovered a number that should trouble employers more than any other: only 15% of respondents have received any AI training at all. This means that the vast majority of German workers have had no formal instruction in how to use AI tools, how to evaluate their outputs, or how to protect sensitive data while using them.
This is not just a problem for individual employees. It is a problem for companies that want to use AI effectively. An untrained workforce is more likely to make mistakes, to fall for misinformation generated by AI, and to be less able to integrate the technology into their daily workflows. It also means that the fear of job loss is probably worse than it needs to be, because workers are left to imagine the worst without any concrete guidance about their own roles.
The AI Act and the literacy obligation
Europe has had a rule about AI literacy for 18 months. Article 4 of the European Union's AI Act has been in force since 2 February 2025. The provision obliges providers and deployers of AI systems to take measures supporting AI literacy among the staff who operate them. The measures must be weighed against the technical knowledge, experience, education, and training of the staff, as well as the context in which the AI system is used.
The AI Act does not set a specific number of hours of training or a particular curriculum. It also binds only organizations that actually use AI systems. This means that the 15% figure is not necessarily a compliance rate. A company that does not deploy AI is not required to train anyone. But for the organizations that do use AI, the expectation is clear: employees should understand how the system works, what its limitations are, and how to use it responsibly.
By any reasonable measure, the state of AI literacy in German workplaces remains weak. Eighteen months after the duty came into effect, most workers have had nothing in the way of training. The gap between the in-force regulation and everyday practice is enormous. This is not just a German problem. It is likely to be a European problem, though representative data from other countries is less readily available.
What employers should take away
The TU Darmstadt study offers a clear message for employers. Workers do not need reassurance alone. They need training and honest answers. Reassuring employees that AI will not replace them is not credible when many of them know better. The better response is to help workers understand what AI can and cannot do, and to invest in reskilling and upskilling so that people can work alongside the technology rather than be replaced by it.
Buxmann's point is not that the future is hopeless. It is that losing a job costs more than just income. It costs identity, social connection, and a sense of purpose. Employers who ignore the training gap are not only failing to comply with the spirit of the AI Act. They are also ignoring one of the clearest warning signs in the labor market. When the people who understand AI best fear it most, the only rational response is to take the fear seriously and do something about it.