AI Transparency: Governance Compliance or Stakeholder Requirements?

Compliance doesn't equal transparency: The paper analyzed 92 AI transparency statements from Australian Government agencies and found that while organizations are meeting formal disclosure requirements, this doesn't necessarily mean the information actually serves the people most affected by these AI systems.

Different stakeholders need different information: Using their RCIN framework, the researchers identified that high-control stakeholders (like government officials) consistently receive the transparency information they need, but high-risk, low-control stakeholders (like citizens affected by AI decisions) receive less substantive information about the systems that impact them.

The "Transparency Illusion" problem: The paper introduces this concept to describe situations where organizations appear transparent on paper through compliance documents, but the actual transparency is unevenly distributed—creating a false sense that stakeholder concerns are being addressed when they may not be.

Mandates alone are insufficient: Simply requiring organizations to publish AI statements doesn't guarantee those statements will address the real information needs of vulnerable populations; governance mandates need to be designed with specific stakeholder groups in mind.

Risk exposure should drive transparency requirements: The research suggests that transparency obligations should be calibrated based on who bears the risk from AI decisions, not just who controls them—meaning those most vulnerable to AI system failures should receive the most detailed and accessible information.