AI Platform Liability for Defamation in Australia: Publisher Status, Stage 2 Reforms, and Emerging Legal Uncertainties

When a machine “confidently" creates a crime you never committed, Australian defamation law is invited to consider who, if anyone, has actually “spoken.” This paper considers the emerging question of whether GenAI platforms can be held liable for defamation under Australian law, focusing on the twin challenges of publisher status and the serious harm threshold in the era of generative AI. It argues that the interaction between strict publication liability and the threshold test creates doctrinal and evidential challenges when defamatory content is generated by large language models rather than human defendants. In jurisdictions that have implemented the Model Defamation Provisions, plaintiffs must show that an AI mediated publication “has caused, or is likely to cause, serious harm” to reputation, displacing any presumption that defamatory meaning itself suffices. For outputs that are ephemeral, personalised and probabilistic, establishing a causal link between specific prompts, particular publications and concrete reputational damage is far from straightforward, particularly where audience size, content stability and republication are uncertain.
The paper situates these challenges within the Stage 2 defamation reforms, considering the narrow exemptions and the new defences for digital intermediaries. It contends that generative AI platforms sit uneasily within this framework, arguing that systems that autonomously generate text look more like originators than passive conduits, limiting access to both innocent dissemination and the new intermediary focused protections. At the same time, the serious harm threshold provides platforms with an under-examined line of defence, that is, many AI “hallucinations” may be low impact, one-to-one communications that may not lead to a serious reputational injury, especially if they are quickly corrected through prompt validation or not republished. Using the aborted Brian Hood/OpenAI dispute and comparative case law, the paper maps the theoretical pathways to liability for platforms and users, and considers how to distinguish AI misstatements from systemic defamatory outputs that warrant full adjudication, with a view to balancing innovation, accountability and effective remedies.