If the only version you heard was the loud one, Canberra had just banned chatbots that act like friends. The version you can pin to the public record is less cinematic and more useful. On 28 October 2026 the Australian Greens said Labor’s Digital Duty of Care Bill had passed the House of Representatives and was headed to a Senate inquiry, with hearings on 9 and 10 November. What ABC News had set out the day before was a duty to take reasonable steps, an age line on algorithmic feeds, and an upper house that will not wave this through.
For anyone putting automated systems in front of customers, the messier version is the one that belongs in the brief. A ban you can screenshot is easy to explain. A duty you have to operationalise is how regulation arrives in a product roadmap.
Reasonable steps, not a blanket prohibition
Clare Armstrong’s ABC News report on 27 October 2026 described the core obligation. Platforms such as Facebook, TikTok and Instagram would have to take reasonable steps to prevent a range of harms for children, including bullying, misogyny and content that promotes eating disorders. For adults, the duty is to limit exposure to serious harms such as threats of violence, stalking and the non-consensual sharing of intimate images.
That is an age split, not one rule for every user and every post. It is also a prevention duty, not a catalogue of forbidden phrases. “Reasonable steps” is how a parliament tells a platform it owns the risk, then leaves the fight over what “reasonable” means to regulators, lawyers and, eventually, the people who ship the product.
Secondary commentary has asserted something sharper: an accepted amendment forcing platforms to ban chatbot material that simulates emotional intimacy and fosters a child’s dependence on it instead of human relationships. That wording is not in the ABC report or the Greens’ 28 October release, the firmest public accounts of the House passage. Until amendment text is on the record, a companion-chatbot ban is unconfirmed. Do not brief it as the law you are about to live under.
The confirmed duty, as ABC described it, lands first on large social platforms, not on every firm that has bolted a chatbot to a website. Pretending otherwise is how vendors sell panic. Ignoring the direction is how operators get surprised. What travels beyond those platforms is narrower: age-split risk, a ranking off-switch, and a complaint path that does not die in a bot.
A feed a 16-year-old can refuse
ABC reported one product change with a number on it. Australians aged 16 and over could switch off algorithm-driven feeds and see only content from accounts they choose to follow.
Those two age lines do not explain each other. Children are covered by a prevention duty. The off-switch, as reported, starts at 16. The reporting does not say that under-16s get the same control, that a separate age rule already keeps them off these services, or that they are deliberately left inside a ranked feed. On the public record so far, the bill draws the switch at 16 and is silent on what ranking control, if any, sits underneath that line. The child duty is a different instrument: prevent specified harms, rather than hand over an off-switch.
For the models that decide what appears next, that 16-and-over setting is not a branding exercise. Ranking that keeps the session going stops being an invisible default and becomes something a user can refuse. Followed accounts only. No suggested pull toward a longer visit.
The Greens want that instinct pushed further. Senator Sarah Hanson-Young has said the party will seek a right for people harmed online to seek compensation, action on what she calls toxic, addictive algorithms, and fines tied to global revenue. Companies, she has said, “only respond to one thing: penalties that hit them where it hurts.” No penalty schedule from the current bill is confirmed in the reporting. Do not invent a dollar figure, and do not treat a Greens demand as a clause already in the text.
The Greens have also said the eSafety Commissioner joined calls for a private right of action the day before their 28 October release. That is their account, not a statement in the commissioner’s own words.
Where slogans have to become duties
ABC reported that Labor had the numbers to pass the lower house without entertaining changes, and that the government needed the Greens to move a duty of care through the Senate the following month. The Coalition party room had already resolved to oppose the bill, after likening it to political censorship.
The Greens support the concept. They are not offering a blank cheque. After House passage they were still demanding “an external complaints pathway, free from the profit driven whims of Big Tech’s content moderation.” Hearings on 9 and 10 November were the dates those sources named. Passage itself was conditional on what the Greens could still extract. A hope of landing the bill by the end of 2026 is a looser timetable than the reporting supports.
The censorship charge deserves a straight reading. A duty aimed at misogyny, bullying and eating-disorder content can be pulled toward ordinary political argument if “harm” is drawn loosely and enforced in the dark. Supporters answer that platforms already rank, suppress and amplify speech every hour, and that a duty stops them pretending the ranking is weather. The test a risk committee can use is operational: who writes the harm definition, is that definition public, and is there an appeal that is not the platform grading its own homework.
A lawsuit is not an ombudsman
Independent MP Sophie Scamps was set to move an amendment for a free, industry-funded online services ombudsman inside the existing Telecommunications Industry Ombudsman. ABC described the cases in plain terms: a small business locked out of a social account, or a person trying to get harmful content addressed. Independents and the Greens favoured it. Labor argued that the draft already required big tech to provide effective internal complaint paths, and had the votes to proceed without entertaining the change. These accounts do not record a formal rejection. They do show the pathway was still an open demand after the House had moved on.
Scamps told ABC that dealing with platforms can feel like “shouting into the void”: automated responses, endless forms, or no meaningful reply. When platforms fail to act, she said, there is nowhere independent to turn. Telecommunications Industry Ombudsman Cynthia Gebert called the amendment “the missing piece” and said the office stands ready to expand. She also said thousands of people and small businesses were already contacting the TIO for help it cannot provide. No exact figure was published, so “thousands” stays a description, not a slide statistic.
Keep that appeal desk separate from the Greens’ courtroom idea. A private right to seek compensation is a lawsuit. An ombudsman, on this proposal, is an outside complaints body funded by the industry it would cover. They answer different failures. Folding them into one “watchdog that pays you” story is how a board memo describes a remedy the bill does not contain.
What should change a roadmap
Three questions travel even if your brand is not on the first list of platforms.
Can the product tell a child from an adult well enough to show it took different care for different risks? A policy PDF does not answer that. Age-split harm is an identity and design problem.
Where is the off-switch on automated ranking? The 16-and-over feed rule is a social-platform measure. The question it teaches is portable. If a system decides what a buyer, employee or citizen sees next (a recommendation, a nudge, a “you might also want”), the refusal has to be a real setting, not a buried preference.
Does a complaint reach a person? Internal tickets that dead-end are already a political exhibit in this debate. If revenue-linked fines or a private right to sue land in the Senate, a bad automated decision stops being a trust-and-safety anecdote and starts sitting next to legal reserves. An agent that refuses, misroutes or silently drops a complaint inherits that exposure. That is a product choice. A slogan will not cover it.
Companion-style agents sit in the blur even without a confirmed intimacy ban. A product built to feel like a relationship generates a duty-of-care argument its founders did not put in the pitch deck. Dependence, especially where children are users, is already a foreseeable harm category in public debate, whether or not this statute names it. Designing for dependence and then pleading surprise is a predictable sequence. It does not qualify as a strategy.
Over-broad duties produce over-blocking. Teams strip useful features because nobody can explain “reasonable” to an auditor. Under-enforced duties produce theatre: the platform marks its own homework, the user gets a polite automated reply, and the harm sits there. Australia has not chosen which failure it prefers. The hearings on 9 and 10 November test that choice. They do not end it.
Questions worth taking to the next briefing
- Has the Digital Duty of Care Bill become law? No. The Greens confirmed House passage on 28 October 2026, with Senate hearings set for 9 and 10 November. ABC reported that Labor needed the Greens to get a duty of care through the upper house, and that the Coalition party room had already resolved to oppose the bill after likening it to political censorship.
- Does the bill ban AI chatbots that simulate emotional intimacy? Not on the evidence that confirms the late-October parliamentary picture. ABC described a reasonable-steps duty with different harm lists for children and adults, not a companion-chatbot prohibition. Secondary commentary has asserted a specific intimacy ban. Treat that as unverified until the amendment text is public.
- Who can turn an algorithmic feed off? As ABC reported the bill, Australians aged 16 and over could switch off algorithm-driven feeds and see only content from accounts they follow. The same reporting does not say what ranking control, if any, under-16s would have. The child obligation is a prevention duty, not an off-switch.
- Is there an independent place to complain? Not as an agreed part of the House bill. Scamps’ industry-funded ombudsman was a live demand. Labor had the numbers to proceed without it, and the Greens were still pushing an external pathway after passage. A private right to seek compensation is a separate Greens demand. Those are not the same mechanism.
- What would “teeth” look like if the Greens get their way? Hanson-Young has tied support to compensation rights for people harmed online, tougher fines linked to global revenue, and tighter control of addictive ranking. No penalty schedule from the current bill is confirmed in the reporting, so do not invent a dollar figure.
- Should a company outside social media rewrite its roadmap as if this statute already applies? No. Meta’s duty is not automatically yours, and the Senate has not made any of this law. The portable lesson is narrower. If your systems rank what people see, talk to customers, or automate the reply when something goes wrong, age, an off-switch and a complaint that reaches a human are the questions this fight is already putting on the table. Treat them as design questions while the statute is contested, not as a legal duty that has already landed on you.
The House has moved a duty. The Senate has not made it law. If you are shipping systems that converse, rank and reply, the useful brief is not that Australia banned friendly chatbots. Reasonable steps, a feed a 16-year-old can refuse, and a complaint that does not vanish into automation are the terms being argued. Design with those questions in view. Do not build as if the argument is already statute.