As artificial intelligence (AI) is rapidly transforming the landscape of education, its influence on assessment practices has become both transformative and controversial. This paper analyzes the potential benefits and concerns of AI-driven educational assessment, with particular attention to the ethical, pedagogical, and social implications of its widespread adoption. Emerging technologies such as Large Language Models (LLM) and machine learning algorithms are enabling new efficient forms of feedback, personalization, and scalability that were unimaginable a few years ago. Nevertheless, many unresolved challenges should be considered, such as algorithmic bias, academic integrity, or legal issues. It is necessary an urgent reframe of AI based assessment through a more ethical and human-centered perspective. By foregrounding values such as fairness, transparency, AI literacy, or inclusivity among other aspects, this work outlines a roadmap for integrating AI tools in ways that enrich and improve assessment experiences, avoiding at the same time drawbacks. The paper proposes ideas to design future AI assessment systems that are both innovative and deeply responsible for students, educators, technologists, and policymakers, finding an ethical perspective for future deployment of smart assessment technologies.