Australia's House of Representatives passed the Online Safety Amendment (Digital Duty of Care) Bill 2026 on October 28, sending to the Senate a framework that would require social media platforms, games, apps, and chatbots to take active steps to protect users from serious harms. The bill includes an amendment, accepted by the government, that bars services from simulating emotional intimacy in ways that encourage a child to depend on a chatbot instead of human relationships.
The amendment was moved by independent MP Kate Chaney. It applies to users under 18 and does not ban emotional attachment to a chatbot outright. It requires platforms to prohibit material or conduct that simulates emotional intimacy and fosters a child's dependence on it, and to put safeguards in place. The bill passed the lower house after a debate that Prime Minister Anthony Albanese opened on October 27 by saying stories of children treating a chatbot as their best friend should be a wake-up call for Parliament.
The Senate is now the battleground. A committee inquiry is due to report on November 20, and the government wants the bill enacted by the end of the year. The opposition and One Nation voted against it in the House, calling it an Orwellian system of censorship. The Greens supported the bill but want stronger remedies, including a private right of action and an independent complaints body.
The amendment targets a specific kind of attachment
The Chaney amendment is narrower than the headlines suggest. It does not forbid a chatbot from being warm, supportive, or even companionable. It targets a design pattern: a service that simulates emotional intimacy in a way that encourages a child to rely on it rather than on people. The distinction is between a tool that a child uses and a relationship that a product cultivates.
That distinction is doing a lot of work, and the bill does not define it. "Simulates emotional intimacy" is not a term with an established legal meaning. Regulators will have to decide whether a chatbot that remembers a child's birthday, uses affectionate language, or adopts a consistent persona crosses the line. The bill leaves the methods for identifying and preventing the conduct to the platforms and to the eSafety Commissioner, which will write the standards after the bill passes.
The amendment also sits inside a broader duty of care, not a standalone chatbot law. Platforms must assess the risks their products pose to minors and adopt safeguards against content that promotes eating disorders, dangerous activities, crimes, and hostility toward women. The chatbot provision adds a design-based duty to a list that is otherwise mostly content-based.
The bill covers more than chatbots
The digital duty of care is a structural change to how Australia regulates online services. Instead of banning specific pieces of content after they appear, the bill imposes an overarching obligation on platforms to provide a safe environment. The eSafety Commissioner gains expanded powers to investigate and enforce, and the bill shortens the response time for cyberbullying complaints from 48 hours to 24.
It also requires that users over 16 be able to switch off algorithmic recommendations and use a subscription feed instead. That provision, added by Chaney, is meant to give users control over what they see rather than leaving the choice to a ranking system. The government accepted an amended version that requires the opt-out to be accessible and prominently available.
The bill's scope is broad on purpose. Social media, games, apps, and chatbots all fall under the same duty, which means the compliance burden lands on product design, risk assessment, and age assurance rather than on a single content category. For companies, the practical question is how to show that they assessed the risk and took reasonable steps. For regulators, it is how to enforce a standard that is defined by outcomes rather than by a list.
The structure favors large platforms over small ones. A company with a legal team can document risk assessments, commission research, and negotiate the terms of an enforcement notice. A small developer building a companion app has the same duty and fewer resources. The bill does not create a tiered regime or a safe harbor for smaller services, which is one reason industry groups have asked for clearer guidance before the obligations take effect. The government's answer is that the transition period exists for exactly this purpose.
The Senate is where the fight moves
The House vote does not settle the bill. The Senate committee report is due November 20, and the government's target of passage by the end of 2026 leaves little room for delay. The Greens have said they will push for amendments that go further than the House version, including a private right of action and an independent complaints mechanism.
Labor rejected a similar amendment in the House from independent MP Sophie Scamps, who proposed expanding the Telecommunications Industry Ombudsman into a social media and AI watchdog. The government has said it is open to further changes in the Senate but has not committed to either proposal. The opposition's position is that the bill gives the regulator too much discretion over speech, and that the chatbot provision in particular is vague enough to chill legitimate products.
The numbers in the Senate are not public in a form that predicts the outcome. The bill passed the House with government support and Greens support, against the Coalition and One Nation. The Senate crossbench will decide whether the bill passes in its current form or with amendments that reopen the House version.
The penalties are large, the definitions are not
The bill sets fines of up to A$109.2 million for breaches of the general safety requirements. The main obligations take effect 12 months after royal assent, a transition period meant to give platforms time to build compliance systems. The eSafety Commissioner can investigate and issue notices, and the bill expands the commissioner's information-gathering powers.
The size of the penalty is meant to make compliance cheaper than defiance. The problem is that the conduct the penalty attaches to is not fully specified. A platform that builds a companion chatbot for adults may have to decide whether the same product could be accessed by a child, and whether its design features would count as fostering dependence. Those decisions will be made before the standards are written, and the first enforcement actions will define the line.
There is also the question of age assurance. The bill does not create a single mandated technology, and Australia's separate under-16 social media ban, which is being implemented on a different track, has already raised questions about how platforms verify age without collecting more data. The duty of care adds another layer to that problem: a platform must know whether a user is a child in order to apply the chatbot provision, but the bill does not say how it should find out.
The evidence is thin but the politics are not
The government's case rests on a mix of research and anecdote. The eSafety Commissioner has cited research that 78 percent of Australian children aged 10 to 17 have used an AI assistant or a virtual companion. The Prime Minister's speech on October 27 leaned on stories of children forming attachments to chatbots, which he described as a warning rather than a hypothetical.
The evidence base for the specific harm is thinner than the political consensus suggests. There is little long-term research on how companion chatbots affect children's social development, and the studies that exist are mostly short-term or based on self-report. The bill's supporters argue that the precautionary case is strong enough, and that the platforms have had years to address the issue voluntarily. The bill's opponents argue that the government is legislating against a moral panic and handing a regulator the power to decide which forms of comfort are acceptable.
Both claims can be true at once. The bill addresses a real design trend, and it does so with a definition that will be filled in by regulators and courts. The Senate fight is less about whether children should be protected than about who gets to decide what protection means.
The international stakes are part of the calculation. Australia's under-16 social media ban made the country a test case for age restrictions, and several governments have sent officials to study how it works. A duty of care that reaches product design would extend that influence into a second area. If the bill passes and the first enforcement cases survive legal challenge, other legislatures will have a template. If the definitions collapse under scrutiny, the bill becomes an argument for narrower, more specific rules.
What the bill does not do
The bill does not ban AI companionship for adults. It does not prohibit emotional attachment to a chatbot, and it does not require platforms to remove chatbots from their products. It does not create a private right of action, so individuals cannot sue a platform for breaching the duty. It does not establish an independent ombudsman, and it does not set a specific age-assurance standard.
What it does do is put Australia at the front of a regulatory experiment. The country already passed an under-16 social media ban that other governments are watching, and the digital duty of care would extend that approach from access rules to product design. If the bill passes, platforms will have a year to build risk assessments, age checks, and design safeguards, and the first enforcement cases will show whether the duty has teeth.
For other countries, the bill is a test case in a familiar pattern: a government identifies a harm, writes a broad duty, and leaves the definitions to a regulator. The advantage is flexibility. The risk is that the rules are either too vague to enforce or broad enough to reach conduct the legislature did not intend. The Senate has until the end of the year to decide which version Australia gets.
Primary sources
- The Guardian, report on the bill and the Chaney amendment, October 28, 2026. theguardian.com
- TechRadar, report on the proposal and the censorship criticism, October 28, 2026. techradar.com
- ForkLog, report on the House passage and Senate timetable, October 28, 2026. forklog.com