Generative Artificial Intelligence (GAI) offers both great opportunities and challenges in education. On one hand, it may provide personalized and interactive pedagogical content that could favor students' intrinsic motivation and active engagement. This may empower them to have more control over their learning, and access diverse knowledge in diverse cultural backgrounds. On the other hand, GAI properties such as lack of uncertainty signalling, low reliability, and steerability could lead to opposite effects, e.g. over-estimation of one's own competencies, persistence in inadequate beliefs, passiveness, impaired curiosity and critical-thinking. These negative effects are amplified by the lack of a pedagogical stance in these models' behaviors. Indeed, as opposed to standard pedagogical activities, GAI systems are often designed to answers users' inquiries in ways that aim to please them, without asking to make efforts, and without considering their learning process. This article outlines some of these opportunities and challenges, with a focus on students' active learning strategies and related metacognitive skills when they use GAI in an educational context. To leverage opportunities and mitigate challenges, we present a framework introducing pedagogical transparency in GAI-based educational applications. This includes 1) methods for training models that consider pedagogical principles; 2) methods to ensure controlled and pedagogically-relevant interactions when designing activities with GAI, involving teachers and other stakeholders and 3) educational methods enabling students and teachers to acquire the relevant skills to properly benefit from the use of GAI in education (meta-cognitive skills, GAI literacy).