TRACEABILITY AND AUDITABILITY IN DECISION SUPPORT: ASSISTIVE COGNITIVE SYSTEMS FOR PLANNING AND GOVERNANCE IN THE PUBLIC SECTOR

Abstract

Introduction: The incorporation of artificial intelligence tools into the Brazilian public administration has advanced in activities such as document analysis, workflow organization, demand triage, and preliminary drafting of administrative documents. In this context, the use of assistive cognitive systems equipped with persistent memory, contextual retrieval, interaction logging, and functional task chaining becomes increasingly relevant, as their operation goes beyond isolated automation and begins to influence procedural and pre-decisional stages. Objective: To analyze the legal parameters applicable to the use of assistive cognitive systems in decision support, planning, and governance within the public sector, with emphasis on traceability, auditability, human oversight, data protection, administrative reasoning, and accountability. Methods: A qualitative, doctrinal legal research approach was adopted, with descriptive-analytical character, based on the examination of the Brazilian Constitution, administrative and digital legislation, particularly norms related to administrative procedure, access to information, data protection, digital government, public procurement, and information security, as well as institutional documents and recent frameworks on artificial intelligence governance in the public sector. Results: It was found that assistive cognitive systems with memory and contextual retrieval require a more refined legal treatment than simple chatbots or isolated automations, as they affect the structuring of internal workflows, the recovery of institutional memory, the drafting of administrative documents, the prioritization of demands, and decision-support processes. Their legitimate use requires functional delimitation, audit trails, verifiable logging of relevant interactions, effective human review, compliance with public purpose, protection of personal data, information security safeguards, and clear mechanisms of responsibility attribution, in order to prevent undue cognitive delegation. Conclusion: The use of assistive cognitive systems in public administration is legally feasible, but depends on robust governance arrangements capable of preserving human authorship in decision-making, ensuring procedural traceability, and aligning technological innovation with legality, transparency, control, and accountability. The study thus proposes foundations for a more precise legal-institutional framework for decision support mediated by auditable cognitive architectures in the public sector.