This paper delves into the formation of a secure and ethical artificial intelligence (AI) framework for the Brazilian public sector in order to propose guidelines to structure innovative models of AI development. Collaborative efforts involving the Public Defender’s Office of Rio de Janeiro, civil society groups, academics, development technology companies, and the Institute for Technology & Society of Rio de Janeiro (ITS Rio) have laid the groundwork testing an inclusive AI development model, addressing the needs of marginalised communities and vulnerable groups. Leveraging Brazil's extensive legal data landscape and the pioneering spirit of public defenders in adopting digital tools, the study employed machine learning techniques to amplify the analysis of judicial data and propose mechanisms to develop an AI for public policy. Through enhanced data analysis, even simple AI solutions can offer profound insights and operational efficiency. The paper highlights the successful implementation of an Operational AI Sandbox approach, ensuring the responsible development of technology in the public sector. It showcases how challenges in terms of participation, representation, and ethical risk mitigation were addressed and expands on how the same model can be applied in other situations. Specifically, the development model used a Multistakeholder Committee strategy encompassing diverse perspectives, to establish ethical guidelines and principles for AI tool development and to monitor and test its actual implementation. This article shares insights gained from the collaborative process, presenting a potential AI model for the public sector. By examining the DPRJ experience, the study shares its step-by-step approach and key takeaways.