Ethical detection of undeclared AI use in academic writing in Spanish

This study investigates the ethical detection of undeclared Artificial Intelligence (AI) use in Spanish-language academic writing using a mixed-methods design. Quantitative results reveal that AI detectors have limited accuracy (85.2\%) and a high false positive rate (18.5\%), performing particularly poorly with hybrid texts (44.3\% accuracy) and showing bias against complex academic Spanish. Qualitative findings identify an "ethical trilemma": institutions face tensions between maintaining integrity, ensuring procedural justice, and preserving pedagogical trust. Document analysis shows an emerging consensus toward "regulated transparency" (56.7\% of policies), where explicit declaration of AI use is prioritized over punitive detection. The study concludes that purely technical solutions are insufficient and potentially unfair. Instead, it proposes a paradigm based on: 1) required declaration policies, 2) critical digital literacy teaching AI's limits, 3) redesigning assessments toward process-based models, and 4) developing Spanish-specific detectors with algorithmic audit systems. The future of academic integrity in the AI era requires less technological surveillance and more investment in intelligent pedagogies and ethical frameworks centered on fundamental human values.