Organizational Perspectives on Mandating AI Transparency

<p dir="ltr">The integration of Artificial Intelligence (AI) into organizational practices brings both transformative potential and significant socio-technical challenges - particularly regarding transparency, governance, and ethics. As governments worldwide move toward mandating AI transparency through regulation, understanding how organizations perceive and prepare for these requirements has become a critical research agenda in Information Systems (IS).</p><p dir="ltr">This study aims to advance the emerging research on AI transparency in IS by focusing on the following research question: <b>How do organizations conceptualize and operationalize mandatory AI transparency in response to proposed regulations? </b>To answer this research question, this study adopts an approach combining manual inductive thematic analysis with Large Language Model (LLM)-assisted Computational Text Analysis (CTA), adapted from previous research. Thematic analysis is executed by analyzing 247 organization submissions to the Australian Government’s “Safe and Responsible AI discussion paper”.</p><p dir="ltr">Thematic findings revealed widespread organizational support for AI transparency, especially when implemented through <b>risk-based approach</b>, <b>audience-specific disclosure</b>, and <b>lifecycle documentation practices</b>. Key differences emerged between sectors - particularly in what information should be disclosed, to whom, and for what purpose. Based on the key themes identified, this study proposes a five-dimensional framework of AI transparency - comprising <b>information source</b>, <b>recipient</b>, <b>content</b>, <b>contextual risk</b>, and <b>purpose</b>. This framework captures recurring concerns across submissions and offers a way to analyze or design AI transparency requirements in practice.</p><p dir="ltr">Methodologically, the study demonstrates that LLM-assisted thematic analysis, when guided by domain-specific prompts, can generate themes that broadly align with human-coded analysis while uncovering additional themes. LLM-generated themes were more topic-centered and concise, whereas manual analysis yielded deeper, action-oriented insights. This contrast underscores the value of a <b>hybrid approach </b>that balances the breadth and scalability of LLMs with the reflexivity and contextual depth of human interpretation.</p><p dir="ltr">This research makes two core contributions: (1) advancing IS understanding of how organizations conceptualize and plan for mandatory AI transparency, and (2) demonstrating a scalable, LLM-enhanced qualitative method for policy analysis. The study concludes that for transparency mandates to be effective, they must be context-aware, risk-sensitive, and aligned with international governance frameworks.</p>