Australia Offers a Feed Switch and a Duty on Chatbots

Australia will force social media platforms to ask users over 16 which feed they want, algorithm or following, under draft laws released on 8 September 2026. The same bill would put a duty of care on games, AI chatbots and other digital services, with fines up to A$109.2 million (US$78.6 million).

Prime Minister Anthony Albanese and Communications Minister Anika Wells framed the package as choice for adults and basic safety standards for products children already use. Platforms have not yet had to build that choice as a default question, and the under-16 account ban that began on 10 December 2025 is still leaking.

Platforms Must Ask Before They Rank Your Feed

The government is calling the feed rule My Feed, My Way. Social media companies would have to send a notification to new and existing users asking what their default primary feed should be.

One path is personalised content ranked by an algorithm. The other is posts from friends and creators the user already follows. Wells told reporters in Canberra the prompt would sit in the open, not in a buried menu, and that people could change their minds as often as they like.

Albanese said the point was control in the user’s hands, not the government’s. “If they don’t follow our laws, they will face significant penalties of more than $100 million,” he said, pointing to the A$109.2 million ceiling in the draft. The exposure draft of the bill is the Online Safety Amendment (Digital Duty of Care) Bill 2026. Feedback is due by 12pm on 22 September 2026, and the government says it will take the bill to Parliament in 2026.

Just as we have basic safety standards for cars, toys or food, the duty of care applies basic standards for the online products we use every day.

Anika Wells, Minister for Communications, Parliament House, Canberra

Wells would not call the model opt-in or opt-out. She called it a “neutral choice” in the middle: algorithm or following feed, offered as the default question when someone opens the app. A journalist asked why the government had not forced an opt-in, meaning the algorithm would stay off until a user turned it on. She stuck with the middle path.

Meta and TikTok declined to comment. Alphabet did not immediately respond.

Europe Already Offered a Switch, Then Buried It

Wells said Canberra had been watching the European Union. Very large platforms there already have to offer at least one ranking option not based on profiling under Article 38 of the Digital Services Act. In practice that option has often sat in settings, reset when an app is reopened, or delivered a thin chronological dump that users abandon.

On 2 October 2025, the Amsterdam District Court told Meta Ireland that Facebook and Instagram had to make the non-profiled option easy to find for each content type and keep the choice from snapping back to a profiled feed. The complaint, brought by Bits of Freedom, treated the reset as a dark pattern.

Wells said Australia’s version would look different because of the pop-up: what do you want as your primary feed, the algorithm or the people you follow? That is a design bet. If the following feed is slow, empty or stripped of features, people will tap back to ranking, and platforms will say the market has spoken.

EU RULES AND THE AUSTRALIAN DRAFT

Researchers at the DSA Observatory have argued that alternative feeds on large platforms are hard to find and poorly understood, so few people stay on them. Australia is trying to solve that with a louder prompt. It is not trying to make ranking illegal.

The Duty Covers Chatbots, Games and Phones

The feed question is the line that travels. The duty of care is the part that changes who is on the hook. Wells said the duty goes past social media so children are protected on chatbots, games, apps and messaging. A May design paper from the communications department said the duty would reach services already under the Online Safety Act 2021, including chatbots, search engines and app stores, plus hosting, internet providers, and equipment and operating systems such as phones and wearables.

Providers would have to keep systems that, so far as reasonably practicable, give Australians a safer service, deal with content that is illegal or harmful to young people, and keep AI and algorithmic recommendation or generation features from causing reasonably foreseeable serious harm. A company would not be fined for one ugly post. Repeated harm, or a failure to fix the system that served it, could be treated as a breach. The eSafety Commissioner would enforce the duty.

Wells used a concrete case from the press conference floor. Girls, she said, have been getting advice from chatbots on how to exist on only 600 calories a day. Boys have been bullied by strangers in games. Young people have copied dangerous or illegal stunts they first saw on a screen. Those examples sit closer to product design than to a following-feed toggle.

Named Harms for People Under 18

For under-18s, digital services would have to screen out design features with negative behavioural effects, including addictive loops and tools that hit self-esteem, and they would have to block a named list of harmful content.

HARMS NAMED IN THE GOVERNMENT RELEASE

• Eating disorders: Content that promotes, encourages or accelerates eating disorders.
• Hostility toward women: Content that promotes hostile ideas about women and gender equality.
• Pornography: Pornography served to people under 18.
• Crime and stunts: Content that glorifies crime or life-endangering stunts.
• Mental health distress: Content that causes serious mental health distress, including abuse and bullying.

Jim Hungerford, chief executive of the Butterfly Foundation, said the helpline hears daily from people whose feeds are packed with idealised, weight-focused posts, which the group treats as a driver of eating disorders. “Every month we lose a hundred Australians to eating disorders,” he said in Canberra, backing the decision to name disordered eating as a harm.

The draft would also let eSafety issue removal notices for nudify apps and websites that generate fake nude images without consent, and it would speed up the child cyber-bullying and adult cyber-abuse schemes. Digital platforms would have to document the steps they take against identified risks and show those steps still work over time.

The Minister for Communications, or eSafety at the minister’s direction, would be able to make binding rules that add categories of harm, set compliance requirements, or exempt classes of services. Those rules would be disallowable instruments and would need a human-rights compatibility statement, including on freedom of expression. That rule-making power is already the part of the package opponents are attacking, more than the feed prompt itself.

Five Million Accounts Came Down, Most Teens Stayed On

Albanese presented the duty as the next step after the social media minimum-age law. “Last year, Australia took world-leading action by banning children under the age of 16 from having social media accounts,” he said. “Our action is working. More than five million accounts have been removed or deactivated.”

eSafety’s own March 2026 compliance update told a harder story. About 70% of children who held accounts on Snapchat, TikTok, Instagram, Facebook and YouTube on the day the ban took effect were still on those apps. The office was investigating those five platforms for possible failures to take reasonable steps. It said it was more satisfied with progress at X, Kick, Reddit, Threads and Twitch, the other five services in the original group of ten.

The regulator had issued 23 information-gathering notices. It described children being prompted to try age checks again, being allowed repeated attempts until a result came back over 16, and finding weak paths to report underage accounts. Some services leaned on a typed-in birthday at sign-up. Julie Inman Grant, the eSafety Commissioner, said in March that the office was moving into an enforcement stance and aimed to decide on court action by mid-2026.

By 8 September 2026, eSafety’s public legal-proceedings list showed a live Federal Court case against Telegram and older matters involving X Corp. It did not list a filed penalty case against Meta, TikTok, Snap or YouTube over the age ban. In June 2026 the government said it would double the maximum penalty for systematic breaches of that ban from A$49.5 million to A$99 million and widen eSafety’s power to pull documents from platforms and from third parties such as age-assurance vendors. Reddit is separately challenging the ban in the High Court. Two teenagers, including 15-year-old Noah Jones, are part of a Digital Freedom Project challenge due to be heard later in 2026.

FROM THE AGE BAN TO THE DUTY OF CARE

1. 10 December 2025: The social media minimum-age ban takes effect, blocking under-16s from holding accounts on ten major platforms.
2. March 2026: eSafety reports about 5 million accounts deactivated, then says 70% of children on five large apps were still active and opens investigations into Snap, TikTok, Facebook, Instagram and YouTube.
3. June 2026: The government says it will lift the age-ban penalty cap to A$99 million and expand eSafety’s information-gathering powers after finding little improvement.
4. 8 September 2026: Albanese and Wells release the Digital Duty of Care draft, adding the feed choice and a systems duty that reaches past social media.

Wayne Holdsworth, whose son Mac died two and a half years before the Canberra event, told the room that platforms already have tools to tell how old children are and choose not to use them. Emma Mason, mother of Tilly Rosemore, who died by suicide in February 2022, said the duty had to look at all technology, not only account bans. Those families were standing behind a second law because the first one has not closed the door.

A Following Feed Still Carries What You Follow

A ranking off-switch does not empty a follow list. Consent campaigner Chanel Contos has spent months on that gap. A 16-year-old who has already followed dozens of gym, hustle and anti-women accounts will, if he picks the following feed, still see those accounts in a row. The algorithm is no longer hunting for the next clip. The library he built is still there.

That is why some safety advocates wanted a true opt-in, with ranking off until a user turns it on, closer to a chronological home screen from a decade ago. Wells refused that default. She said many Australians like the algorithm for local businesses and entertainment, and that the job was to offer the choice and then respect it. Albanese, asked about people who still want to scroll without end, said that was up to them.

Dublin City University researchers built ten sockpuppet accounts mimicking 16- and 18-year-old boys on TikTok and YouTube Shorts. Every account was served toxic content within 23 minutes, and manosphere content within 26 minutes, whether the account had searched for that material or for ordinary male-interest clips. After about 400 videos, or two to three hours, 76% of TikTok recommendations and 78% of YouTube Shorts recommendations were classed as problematic or toxic, mostly alpha-male and anti-feminist clips.

A following feed would cut that hunt for the next recommended clip. It would not wipe accounts already followed, and it would not, on its own, rebuild how a chatbot answers a question about calories or how a game treats a 14-year-old in voice chat. Those jobs sit in the duty, if the duty is written tightly and then enforced.

Henderson Calls the New Powers a Speech Threat

Coalition communications spokeswoman Sarah Henderson has called the proposed powers a “serious threat to free speech.” Opposition Leader Angus Taylor has described the plan as an attempt to censor social media. Nationals leader Matt Canavan said he wanted children protected from what he called radical gender content and doubted that was the government’s main aim.

Independent MP Allegra Spender said she could not see how letting people opt in and out of algorithms was censorship. Wells, asked whether “harm” in the bill could be stretched over political speech or misinformation, said the adult prohibitions were illegal material: child sexual exploitation, grooming, violence, sexual violence, menacing or harassing conduct, and the promotion of terrorism and extremism. “Apart from that, this is about choice,” she said.

Dr Andrew Bushnell, director of the law and freedom program at the Institute of Public Affairs, went further. “The Prime Minister wants to determine what all Australians can do, see, and say on the internet, and shut down political debate, through the proposed Digital Duty of Care,” he said. “Australians should be wary of trading the algorithm for the ‘Alborithm.’” His list of objections starts with a wide definition of serious harm, a fear that dropped misinformation rules could return through duty language, and an incentive for companies to over-block rather than risk a nine-figure fine.

That is the argument that already dominates the public fight. The feed toggle is easy to like in a poll. The minister’s power to name new harms, and a duty that can touch messaging apps and operating systems, is where the bill will be amended or blocked.

THE DRAFT ON THE CLOCK

• Penalty cap: Failure to meet the duty could bring fines up to A$109.2 million, enforced by eSafety.
• Consult close: Comments are due by 12pm on 22 September 2026 to digitaldutyofcare@communications.gov.au.
• Age-ban record: About 5 million accounts have been removed or deactivated; 70% of children on five large apps were still active in March 2026.
• Transition talk: The May design paper proposed 12 months between passage and the duty starting, after industry asked for at least that long.

Albanese said more than 20 countries had moved toward similar age limits and that Australia should keep going. Wells said tech firms could still move fast but would no longer be allowed to break things. The draft now sits with platforms, child-safety groups and a Coalition that has been told by its right flank to treat the whole package as a speech bill. What users will actually see, if the law passes, is a pop-up asking which feed they want, and a second product behind it that has to work well enough that the choice is real.