<![CDATA[This study systematically reviews the credibility of artificial intelligence (AI)–based authentic assessment in evaluating numeracy achievement among early-grade primary students. The growing use of AI in educational assessment—ranging from automated scoring to adaptive feedback—has raised questions regarding its validity, reliability, and fairness. Using a qualitative descriptive approach through a systematic literature review, this study synthesizes eighteen high-quality publications issued between 2020 and 2025 from major academic databases. Thematic analysis identifies three interrelated findings. First, algorithmic accuracy must be complemented by human validation to prevent bias and preserve contextual meaning. Second, collaboration between teachers and AI systems enhances interpretive credibility and fosters reflective learning practices. Third, integrating AI ethics and literacy within primary school curricula is essential to ensure fair and transparent assessment outcomes. The results indicate that credible AI-based evaluation depends not solely on computational precision but also on socially grounded interpretation supported by ethical and pedagogical principles. Theoretically, this study reinforces Vygotsky’s social constructivist perspective and the Assessment for Learning (AfL) paradigm, positioning AI as a cognitive scaffold that enables reflective interaction among students, teachers, and technology. Practically, it offers implications for educational policy in Indonesia’s Merdeka Curriculum, emphasizing teacher training, ethical guidelines, and algorithmic transparency to promote equitable digital learning ecosystems. This study uniquely integrates AI assessment credibility with the context of early numeracy, providing original insights and advancing theoretical discourse on ethical, human-centered, and contextually responsive digital assessment practices in primary education.]]>
