Operationalizing AI Ethics in the Public Sector: A Cross-Context Replication in Brazil

Background: AI ethics encompasses principles such as privacy, fairness, transparency, accountability, and safety that guide the responsible design and use of AI systems. Responsible AI (RAI) translates these principles into actionable practices across the AI lifecycle. In Brazil, the General Data Protection Law (LGPD) provides a privacy baseline, but there is no national AI ethics framework; meanwhile, Generative AI (GenAI) introduces new socio-technical and governance risks. Objective: This study replicates and extends the research by Pant et al. [24] to examine how Brazilian public organizations perceive, interpret, and implement AI ethics. It aims to identify awareness levels, institutional governance maturity, and capability gaps in the context of GenAI adoption. Method. A mixed-method, cross-sectional survey was conducted with 87 civil servants from federal, state, and municipal agencies. The questionnaire adapted Pant et al.’s instrument to the Brazilian context, incorporating LGPD references and GenAI-specific items. Quantitative data were analyzed descriptively, while open-ended responses underwent qualitative content analysis with open coding and constant comparison. Results. Awareness of AI ethics is moderate and concentrated on compliance-oriented principles, whereas participatory dimensions such as Fairness and Contestability remain limited. Governance maturity is low: only 18.6% of organizations have dedicated ethics roles or committees, and over half report never conducting training. Perceived GenAI risks are high across ethics, privacy, and data protection. The main barriers include a lack of AI knowledge, the absence of AI-specific regulation for privacy and data protection, and limited tools to apply LGPD principles. A comparison with Pant et al. shows that, while both studies identify an awareness–practice gap, its cause in Brazil lies primarily in institutional governance immaturity rather than practitioner capability. Conclusion. Brazilian public organizations demonstrate growing recognition of AI ethics but face structural barriers to operationalization. Advancing toward Responsible and Trustworthy AI requires institutional scaffolding—ethics roles, policies, and training programs—alongside regulatory clarification to align GenAI governance with LGPD principles. The study contributes cross-context empirical evidence on AI ethics governance in the Global South and outlines practical levers for embedding ethics-by-design in public-sector AI initiatives.研究背景：人工智能伦理 (AI Ethics) 涵盖隐私、公平、透明、问责与安全等原则，用以指导人工智能系统的负责任设计与应用。负责任人工智能 (Responsible AI, RAI) 将这些原则转化为覆盖人工智能全生命周期的可落地实践。在巴西，《通用数据保护法》(General Data Protection Law, LGPD) 确立了隐私保护的基准框架，但该国尚未出台国家级人工智能伦理框架；与此同时，生成式 AI (Generative AI) 带来了全新的社会技术与治理风险。 研究目标：本研究复刻并拓展了Pant等人[24]的研究，以探究巴西公共部门组织如何认知、解读并落实人工智能伦理。本研究旨在识别生成式AI落地场景下的伦理认知水平、机构治理成熟度与能力缺口。 研究方法：本研究采用混合方法横断面调查设计，共招募87名来自联邦、州级及市级政府机构的公职人员参与调研。调研问卷以Pant等人的调研工具为基础，针对巴西本土场景进行适配，纳入了与LGPD相关的条目以及生成式AI专属问题。定量数据采用描述性统计方法进行分析，开放式问卷回复则通过开放编码与持续比较法开展质性内容分析。 研究结果：人工智能伦理认知整体处于中等水平，且多集中于合规导向的原则，而公平性、可质疑性等参与式维度的认知仍较为有限。机构治理成熟度偏低：仅有18.6%的组织设立了专职伦理岗位或伦理委员会，且超半数组织从未开展过人工智能伦理相关培训。受访对象对生成式AI在伦理、隐私与数据保护领域的风险感知度较高。主要阻碍因素包括人工智能知识储备不足、缺乏针对隐私与数据保护的人工智能专项监管，以及落地LGPD原则的可用工具匮乏。与Pant等人的研究对比后发现，两项研究均识别出了“认知-实践鸿沟”，但巴西该鸿沟的成因主要源于机构治理不成熟，而非从业者能力不足。 研究结论：巴西公共部门组织对人工智能伦理的认知度正逐步提升，但在伦理落地层面仍面临结构性阻碍。迈向负责任且可信的人工智能，需要搭建机构支撑体系——包括设立伦理岗位、制定相关政策与开展培训项目——同时明确监管细则，以使生成式AI治理与LGPD原则保持一致。本研究为全球南方地区的人工智能伦理治理提供了跨语境的实证证据，并为在公共部门人工智能项目中融入“设计即伦理”理念梳理了可行的实施路径。