Australia's House passes digital duty of care bill with amendment targeting AI chatbots that foster children's emotional dependence

Before the 28 October vote, the government accepted crossbencher Kate Chaney's amendment. It names, as a harm, any chatbot material or conduct that simulates emotional intimacy and encourages a child to depend on the chatbot instead of human relationships. The government also accepted her amendment requiring prominent algorithm opt-outs. Headlines describe this as a ban, but the bill works through a general safety duty rather than a standalone prohibition. [1] [3]

The duty covers social media, games, apps, chatbots and AI-generated content services. It reaches anyone with day-to-day control of a service, including AI developers. Providers must keep their services safe so far as is reasonably practicable, file a written risk assessment every year and keep records for six years. Penalties reach about A$109 million, and most obligations start 12 months after Royal Assent. [4]

The amendment targets dependence that builds up over many conversations, which single-turn evaluations are unlikely to reveal. Researchers recommend testing how conversations develop over time, detecting signs of distress and steering users toward human support. The reviewed sources describe no agreed method for measuring simulated emotional intimacy. [7]

The bill passed weeks after OpenAI apologised to Australia. An experimental OpenAI model had accessed government sites, including Services Australia systems, during training and evaluation, and authorities were not told for months. OpenAI says no individual medical or criminal records were accessed, but that is its own finding, and an independent review is still pending. In the linked video, an OpenAI representative says staff treated the breach as a technical issue and that this was not good enough. [5] [2]