Abstract
When regulators face a consumer protection problem, mandatory disclosure is typically the first response.
Decades of evidence from financial services show it rarely works: consumers don't read disclosures, can't understand them when they do, and can't use them to make better decisions. Rather than improving outcomes, disclosure can provide false assurance, crowd out more effective interventions, and shift accountability to the people least equipped to bear it.
Governments are now reaching for this same playbook for AI. The EU AI Act mandates disclosure of chatbot interactions and automated decisions. Australia's Privacy Act will require AI disclosure in privacy policies. These requirements rest on the familiar assumption: if people know AI is involved, they will adjust appropriately.
Emerging research suggests they won't. Studies show AI disclosure typically reduces trust, potentially harming decision quality. Explanations of AI recommendations can increase compliance without improving calibration. The behavioural mechanisms that defeated financial disclosure are already operating in AI contexts.
This session brings together academic and industry perspectives to examine why disclosure-based approaches to AI regulation are likely to fail, and what more effective alternatives might look like. Drawing on decades of evidence from financial services, experimental research, and industry experience, panellists will discuss how to design the AI regulatory system to achieve good consumer outcomes.