Australia moves to ban AI chatbots built for emotional intimacy
Australia's federal government wants to outlaw AI chatbots designed to create emotional intimacy, as a new online safety bill advances and pressure grows for wider oversight.
According to The Guardian, the Australian federal government wants to ban AI chatbots that seek to create "emotional intimacy" with users. The proposal follows a speech by Prime Minister Anthony Albanese, who expressed concern that children are treating these AI agents as their best friend.
The Guardian reports that the idea is linked to a new online safety standards bill, which has passed the lower house of the federal parliament. The Labor government is also facing growing calls to take further steps in the same area.
Those calls, the outlet says, include giving the telecommunications ombudsman the power to act as a watchdog for social media and AI services. Advocates want Australians to have an external complaints mechanism and the ability to publicly name and shame large technology companies.
The reporting does not yet explain how "emotional intimacy" would be defined in law, when any ban would take effect, or how it would be enforced. Those details will matter as the bill moves forward and as lawmakers decide how far the oversight should reach.
The Guardian places the debate within a wider argument about whether technology companies should bear a clearer duty of care toward the people who use their products, particularly young people.
Background
Australia has become one of the most active countries in regulating online platforms. The Online Safety Act 2021 set out the powers of the eSafety Commissioner, who can require platforms to remove certain harmful content and who oversees industry codes covering areas such as child protection. Over the past few years the government has also moved to restrict social media access for younger teenagers, a step that drew both support and criticism from researchers, parents and industry groups.
The phrase "duty of care" describes an obligation on companies to take reasonable steps to prevent foreseeable harm to users. The idea has been adopted in other jurisdictions. The United Kingdom's Online Safety Act 2023, for example, places risk-assessment and safety duties on platforms, with a particular focus on protecting children.
AI companion or chatbot apps are designed to hold ongoing, personal conversations. Many people use them for entertainment, practice or companionship. Some clinicians and researchers have raised concerns that frequent use could foster dependence, blur the line between software and a relationship, or expose vulnerable users to unhelpful advice. The evidence base is still developing, and opinions differ on how serious the risks are and for whom.
Ombudsman schemes are a familiar tool in consumer protection. They provide a free, independent route for people whose complaints a company has not resolved. Extending a similar model to digital platforms would be a structural change. Public naming of companies is another regulatory lever, used by some agencies to create reputational pressure for compliance, though its effectiveness depends on how and when it is applied.
Regulators in several countries are now examining how conversational AI systems interact with minors. Australia's approach may be watched closely because it combines a legislative bill, a possible product-level restriction and a proposed complaints system in a single policy push.
Why it matters
The proposal raises a basic question: where does a helpful tool end and an emotional relationship begin? A chatbot that answers questions is one thing. A system built to simulate closeness, affection or dependence is another, and lawmakers will need a definition precise enough to target the second without sweeping in ordinary software.
The issue is especially sensitive for children, who may not recognize that a responsive system is designed to keep them engaged. Supporters of tighter rules argue that governments should act before patterns of dependence become widespread. Critics are likely to ask whether a ban can be enforced in practice, especially when many products are offered by overseas companies and can be accessed through general-purpose apps.
There is also a question of balance. Adults who find value in AI companionship may object to rules that limit what they can use. Any framework will need to weigh protection against autonomy, and to explain how age checks or other safeguards would work without creating new privacy risks.
Finally, the proposed complaints mechanism and public naming power would change the relationship between regulators and platforms. If an external body can receive complaints and publicize failures, companies would face stronger incentives to respond quickly. How that power is defined will shape how much influence it has over the sector.
What to watch
The first thing to track is the bill's progress through the Senate and any amendments that change how emotional intimacy is defined. Narrow wording could limit the ban to specific design features, while broader language could cover a wider range of products.
Next, watch whether the telecommunications ombudsman's role is expanded to cover social media and AI, and whether the government provides details on funding, staffing and powers. The scope of any naming-and-shaming provision will also be important.
Enforcement is another key area. Penalties, age assurance requirements and the process for handling overseas providers will determine whether the rules have real effect. Industry responses, including changes to product design or access in Australia, could signal how companies plan to comply.
Finally, observers will want to see how other governments respond, since Australia's choices may influence similar debates elsewhere.
Viewpoints
Commentary from our columns: three angles on this story.
The Rulebook privacy, safety and regulation
Australia's proposal asks a sharp question that many regulators have only begun to face. When software is built to feel like a friend, the rules should ask whether that design choice creates foreseeable harm, not only whether the content is unlawful. A duty-of-care approach fits this problem well, because it focuses on how a product behaves rather than on a list of banned words or topics.
The hard part is drafting. "Emotional intimacy" is a concept that needs careful legal definition, or it could catch harmless features or leave loopholes that a clever developer can exploit. Lawmakers should also be explicit about privacy. Age checks and identity verification solve one problem but can create data risks of their own, so any safeguard should be proportionate and tightly limited.
Public complaints channels and the power to name companies can be valuable if they are transparent and fair. Clear standards, published reasons and a right of reply would make the system credible. The aim should be safer design, not a regime that punishes experimentation or leaves users with nowhere to turn.
The User what changes for people who use it
For most people, the change would show up in ordinary moments. A teenager chatting with an app late at night might find that certain features are no longer offered, or that the bot now declines to act as a constant companion. An adult who uses an AI system for comfort might notice new prompts, limits or warnings about how the product works.
Those changes could be useful, but they could also be frustrating if the rules are vague. People who rely on these tools for support deserve to know what is allowed, why a feature has changed and where to go if something goes wrong. A complaints mechanism would help only if it is easy to find and quick to use.
Families also have a role. Rules can set minimum standards, but parents and carers will still need simple ways to understand what apps their children use and how those apps respond. Good policy should make that job easier, not add another layer of paperwork.
The Skeptic open questions and weak points
The headline promise is simple: ban chatbots that create emotional intimacy. The questions underneath are harder. Who decides what counts as intimacy? A companion app, a customer service bot with a friendly tone, and a general assistant that remembers a user's name may sit at very different points on a spectrum, and it is not obvious where a line should go.
There is also the enforcement problem. Many AI services are run by companies outside Australia and can be reached through browsers, open-source tools or other platforms. A ban that applies mainly to large, cooperative providers may simply move demand elsewhere. Policymakers should be asked how they plan to measure compliance and what happens when a service ignores the rules.
Finally, the proposal relies on evidence about harm that is still being gathered. Governments are right to act cautiously in both directions. Before committing to sweeping restrictions, the public deserves clear data about who is affected, how severe the risks are and whether a narrower rule would work better.