The proliferation of Artificial Intelligence (AI) is profoundly transforming educational systems, creating both opportunities and risks. Although research on its impact on performance and inequality in education is expanding, empirical evidence remains scarce. This is particularly true in the field of Vocational Education and Training (VET), a pathway of increasing importance in developed countries with strong potential to contribute to youth employment and economic development.

Our joint study by the Spanish National Research Council (CSIC) and Ayuda en Acción analyzes Spanish evidence to advance three main objectives: (i) to examine how AI is being incorporated into VET, comparing public and private institutions as well as VET and other educational pathways; (ii) to assess its impact in terms of inequalities of access and use; and (iii) to identify the limitations and risks that hinder an inclusive, ethical, and effective integration of AI in VET.

Our research covers Basic, Intermediate, and Advanced levels of VET and combines quantitative evidence from an online survey of 355 teachers and 386 students conducted between April and May 2025 with qualitative insights from eight expert interviews and 15 in-depth interviews with educators and learners. These findings are further contextualised through a meta-analysis of 121 academic publications.

Findings reveal that:
- Generative AI use is significantly higher in public schools (54% of students) than in private ones (39%), highlighting inequalities linked to resources and family support.
- AI integration often reinforces existing divides: usage is higher among students in Advanced VET (62% of students) than in Basic VET (33%); in urban areas (57%) compared to rural ones (34%); among men (56%) versus women (41%); and among students from more affluent households with better access to devices and connectivity.
- Teachers report a severe lack of training: only 12% of teachers and 9% of students have received any formal instruction on AI.
- Awareness of algorithmic bias and ethical risks remains strikingly low: fewer than 15% of respondents (teachers and students) stating that they understand these issues well.

In conclusion, this study shows that, without proper planning, AI in the context of VET tends to reinforce existing inequalities related to education level, geography, age, gender, and access to technology—thereby limiting its potential to nurture talent in vulnerable contexts. The absence of coherent public policies to guide the integration of AI in VET further exacerbates these challenges. Addressing them requires moving beyond a purely instrumental approach and promoting inclusive models that value diversity and critical thinking, while ensuring human oversight to safeguard learning diversity. It is equally vital to raise awareness of the risks associated with irresponsible AI use by strengthening transparency, regulation, and the auditing of algorithmic systems. Ultimately, AI can become a powerful ally in improving VET and contributing to youth employment and social inclusion—provided it is developed with digital justice, social inclusion, and democratic participation at its core, with social and educational organisations playing a pivotal role in advancing human-centred AI.

Keywords: Artificial Intelligence, digital divide, inclusive education, educational equity, higher education.