This means that, when problems arise, it should be possible to trace their causes and to understand the basis, logic, and key factors that influenced an AI system’s decisions. In addition, developers of AI systems have a responsibility to provide deployers with relevant information, and deployers of AI systems, in turn, have a responsibility to provide necessary information to those affected by their use. That said, these principles may not apply in the same way to users of generative AI services in all cases. With generative AI, the reasoning or basis for an output may be embedded in the output itself or may not be particularly relevant. For example, a user can readily understand why a particular image was generated based on the prompt they provided, whereas it may be impossible to explain which specific training data led to the generation of that image. Civil society organizations, which place strong emphasis on the principles of accountability and transparency, need to ensure transparency in their use of generative AI to the greatest extent possible. This is because transparency enables those affected by AI-generated outputs to make informed judgments, thereby strengthening trust in both AI systems and the organizations that use them. For example, if a document is summarized using generative AI and its accuracy may not be complete, this fact should be clearly indicated so that audiences can take it into account when assessing the reliability of the information. In addition, confusion may arise when audiences mistake generative AI outputs for human- 58 59