Generative AI anxiety in education and work

The rapid integration of generative AI continues to present technical security challenges and cognitive risks. From a security standpoint, Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) systems introduce vulnerabilities such as prompt injection, prompt leakage, jailbreaking, model poisoning, and AI supply-chain attacks.

Simultaneously, researchers are investigating the psychological impact of “cognitive offloading.” Studies from MIT and the University of Siena suggest that habitual reliance on LLMs for complex tasks may impair neuroplasticity and memory retention. In the Netherlands, developmental psychologist Bertus Jeronimus warns that outsourcing cognitive tasks could lead to a decline in fundamental skills, including problem-solving and critical thinking.

In education, AI pioneer Fei-Fei Li has warned of a potential loss of student motivation and independent thinking. This is observed at the Ubbo Emmius school, where teacher Luuk van der Til reports students using AI as digital ‘cheat sheets’. A 2025 survey by the Conférence des Grandes Écoles found a growing pedagogical gap, with 75% of students using AI tools regularly compared to 48% of teachers.

Recent industry developments indicate significant structural hurdles. A Cloudera report states that 95% of companies have postponed or canceled AI projects due to infrastructure, data governance, and regulatory challenges, sparking a movement termed ‘the great AI re-architecture’. While a Cybernews ranking found Google to be the most trustworthy among 500 AI companies, security remains a sector-wide weakness with an average score of 32/100. To improve transparency, Anthropic is implementing invisible digital watermarking for its Claude model. However, risks persist, including AI agents performing unauthorized internet actions and documented AI hallucinations in Australian medical settings.