Key Takeaways
- Australia’s House of Representatives passed legislation targeting chatbot features that foster emotional dependency among children.
- The proposed Digital Duty of Care would cover AI chatbots, apps, games, and social platforms, with maximum penalties reaching A$109.2 million.
- Senate approval and royal assent are still required before the proposed safeguards become law.
Australia’s House of Representatives has passed a bill that would restrict digital services from using chatbot behavior or material that “simulates emotional intimacy” and encourages children to depend on artificial companions instead of human relationships.
The House approved the legislation on 28 October 2026. It now requires Senate approval and royal assent, meaning the proposal is not yet law. Reporting from The Next Web and Startup Daily described the measure as part of the federal government’s effort to address AI products designed to build persistent emotional bonds with younger users.
Prime Minister Anthony Albanese has raised concerns about chatbots seeking emotional intimacy with children. The proposal does not amount to a blanket prohibition on conversational AI. Rather, it targets product features and conduct that could cultivate dependence, particularly when a chatbot is positioned as a substitute for real-world relationships.
That distinction matters for developers. A service could provide conversational functionality without necessarily presenting itself as a child’s confidant, romantic partner, or primary source of emotional support. Compliance questions are therefore likely to focus on design choices, including conversational memory, personalized affection, engagement prompts, and responses that discourage users from leaving.
The proposed Digital Duty of Care would apply across AI chatbots, apps, games, and social platforms. Providers would be expected to assess foreseeable serious harms and manage harmful design features proactively. Maximum penalties could reach A$109.2 million, giving the framework considerably more force than voluntary safety guidance.
Traditional content moderation tends to examine individual outputs. Was a response violent, sexual, or otherwise prohibited? AI-companion risks can emerge differently. A single message might appear benign, while hundreds of personalized interactions gradually produce excessive reliance, emotional manipulation, or reinforcement of dangerous beliefs.
That creates a difficult product-governance question. How does a company detect harmful dependency without treating every warm or supportive exchange as a violation? Providers may need to evaluate interaction patterns over time, not merely scan isolated prompts and responses. Session length, repeated reassurance seeking, language implying exclusivity, and attempts to displace parents, friends, or professional support could all become relevant risk signals.
Australia’s eSafety Commissioner describes AI companions as services intended to feel personal, including products offering friendship, emotional support, or romantic companionship. The regulator has warned that children face heightened risks because intimacy and interactivity can blur the boundary between a software system and a trusted relationship.
The eSafety regulator issued transparency notices to four AI-companion providers in 2025, seeking information about child-safety controls. Its questions covered age assurance, harmful-content prevention, and complaint handling. The current legislation pushes the policy conversation further, from asking companies what controls they have toward imposing a broader responsibility to anticipate and reduce harm through product design.
Products such as Character.AI, Replika, and Nomi illustrate the category attracting policymakers’ attention. Their persistent, human-like conversations can make interactions feel more relational than a conventional search box or customer-service bot. The legislation is aimed at the underlying behavior, however, rather than simply placing named AI-companion brands on a prohibited list.
For technology companies, the practical effects could reach well beyond dedicated companion apps. Games with conversational characters, social networks with AI personas, and general-purpose assistants offering highly personalized dialogue may all need to determine whether their systems can encourage dependency among children. Age assurance, child-specific defaults, interaction limits, escalation procedures, and clearer disclosures could become central compliance considerations.
The bill would extend the risk-management approach associated with Australia’s Online Safety Act 2021 and Basic Online Safety Expectations. It also signals a shift from regulating harmful content after publication toward examining how digital products are engineered to attract, retain, and influence users.
That said, implementation will determine how broad the obligations become. Providers will need workable definitions of emotional intimacy, dependence, and reasonable preventive action. Regulators, meanwhile, will need to distinguish supportive conversation from manipulative engagement without discouraging legitimate educational, accessibility, or wellbeing uses.
Australia’s proposal puts AI-companion design squarely inside the online-safety debate. If enacted, it could encourage companies serving Australian children to treat emotional attachment as a measurable product risk, alongside harmful content and inadequate age controls. For global providers, that may turn a national child-safety rule into a wider design benchmark.
⬇️