Riesgos técnicos y responsabilidad jurídica en el uso de la inteligencia artificial en el sector público colombiano

Abstract

The adoption of artificial intelligence (AI) by Colombian public entities raises a legal and technical difficulty: determining who is liable when an automated decision causes harm and the path leading to that decision cannot be clearly reconstructed. This reflection article examines how model opacity, algorithmic bias, and generative AI hallucinations relate to the patrimonial liability regime applicable to the public sector. To this end, a qualitative content analysis was conducted on thirty-one documentary sources, using coding stages inspired by grounded theory, which made it possible to compare technical and legal findings from an interdisciplinary perspective. Although each risk compromises different constitutional guarantees, the results reveal a common pattern: harm may be identifiable even when the information needed to explain its cause and attribute liability is unavailable. Existing duties of assessment, documentation, and transparency do not always translate into verifiable records, institutional controls, or enforceable contractual obligations. The issue, therefore, goes beyond the weak implementation of existing duties and the lack of traceability needed to reconstruct automated decisions, and calls for specific binding regulation. Accordingly, the article proposes strengthening human oversight, record retention, supplier obligations, and the assessment of a liability regime for high-risk systems, as well as enacting a comprehensive statutory law on artificial intelligence.