The introduction of generative intelligence into educational contexts calls for an immediate reconsideration of the risks and ethical responsibilities associated with its use. The focus is on the impact of generative intelligence on the construction of autonomous thinking and the need to preserve inclusive conditions of access and participation. This chapter develops a systemic analysis of the vulnerabilities emerging from the use of generative models in education, articulating a taxonomy that includes ethical, psychological, cultural, social, normative and technological dimensions. From this, guidelines are proposed to steer AI integration practices consistent with the principles of accountability, safety and educational sustainability. Within this framework is Socrates AI, a dialogic agent designed to support critical and reflective learning through adaptive interactions oriented toward personalization, inclusiveness, recognition of individual differences and ethical management of dialogue. Socrates AI builds on its own certified knowledge base, on the maieutic approach, and on effective questioning models, within a transparent and flexible architecture. Preliminary evidence from field deployment suggests its potential to mitigate cognitive risks and foster epistemic responsibility. The chapter presents the theoretical and design foundations of the proposal, contributing to the debate on the conscious use of artificial intelligence in education and citizenship, while opening up the prospect of an experimental evaluation of the cognitive impact of AI through psychopedagogical neuroscientific tools.