Chanel Contos spent her National Press Club address in Canberra on Thursday arguing against the remedy her own campaign made politically available. Fix Our Feeds put algorithmic recommendation on the Australian agenda by naming what it delivers: online misogyny, sexual violence, the normalisation of pornography and of choking during sex. The sexual consent advocate told the club that it is too easy for large technology companies to build schemes that look like structural reform but in practice suit them entirely. User control over what the algorithm serves is not accountability, she said, and it is less so when the user has to switch the system off themselves.
That distinction is the whole argument, and it is about to be tested. An algorithmic opt-out is the kind of measure that can be announced in a sentence, shipped as a settings toggle, and counted as a response.
I read the under-16 social media ban as the template for exactly this failure: simplified, in places unworkable, and easy to announce. The government's effort on gambling advertising has the same shape. So does an opt-out. Each leaves the responsibility sitting with the consumer, which is the cheapest place for a platform to leave it. Opting out of algorithmic recommendation is one component of a system for governing an environment the algorithm has already reorganised, and on its own it governs nothing.
What is being opted out of is not a single piece of software. It is a set of systems that assemble a profile of a user and then use that person's habits to keep them scrolling. The conditions that produces are already visible: features built to form dependency, shortening attention spans and reducing productivity; rising rates of mental health problems; conflict and outrage travelling further than material that supports consent and common ground; and a statistical increase in social isolation across every age group.
The counter-proposal is a legally binding, enforceable duty of care owed by large technology companies to their users — a way of returning some measure of information sovereignty, the ability to affect what happens in a digital environment rather than only receive it. The timing argument is the strongest part of the case. AI is advancing by the minute while governments try to handle developments they do not know how to govern, and in a contest between big technology companies and the public it is not hard to predict who usually wins. Smart glasses are only the latest instance of a difficult debate that is also an opening to rewrite the rules. Updating Australia's long-obsolete privacy laws alongside a statutory duty of care would be a serious step.
The text already exists. In 2024 Zoe Daniel introduced a private member's bill on a digital duty of care in the federal parliament, drafted on the most effective provisions in United Kingdom and European Union law. Its second reading speech put the purpose plainly: to make platforms make their spaces safe for users, children and communities. The duty would cover all the systems, processes and elements of a digital service, including dark patterns — interface choices that push a person toward a particular action — and features engineered to form dependency.
The bill provided for:
the strictest privacy settings by default;
the ability for users to reset the algorithm and to turn it off entirely;
specific measures for young people with eating disorders and mental health problems, people with gambling addiction and other vulnerable groups;
detailed evidence that platforms assess risks and act to reduce them;
transparency of algorithmic operation and access for independent researchers to how these systems work internally;
penalties for non-compliance of up to 10% of a company's global turnover.
Read that list against the opt-out and the asymmetry is obvious. The opt-out appears inside the bill as one line item among six, and it is the only line item a platform can satisfy without changing anything about how it operates. The 10% of global turnover is the part that would actually bind, and it is also the part that makes the bill hard to pass — which is the most reliable indicator anyone has that it is the serious proposal in the room. Notably absent from the opt-out framing is any obligation that survives the user's own choice.
My own doubt about the reset provision is worth stating separately, because it cuts against the bill too. A reset assumes the profile is the durable object. I suspect the durable object is the ranking objective — the thing the system is optimising for — and no user-facing setting touches that. Transparency and researcher access are the provisions that would find out. The reset is the provision that would feel like an answer while the system rebuilt itself around the same target.
The point of a duty of care is that it protects everyone, whether or not a given person has opted out of anything. Platforms would have to answer for how their systems are built and for the risks that design creates — to emotional wellbeing, to political and social cohesion, and through features that form dependency. That is a shift away from the old approach of removing individual pieces of content one at a time, forever reacting to the next violation, and toward responsibility for the machinery that decides what spreads.
False and deliberately misleading information could be folded into such a system, and doing so would immediately summon the familiar free speech fight. The cultural and social changes that Mark Zuckerberg, Elon Musk and other technology executives impose without obstruction are, meanwhile, simply absorbed as the way things are. The real question is sovereignty over the information environment, and whether people govern the systems companies built or those systems govern them.
Daniel's bill has been sitting in the parliament since 2024. Every month it is not the government's answer is a month in which the answer on offer asks the user to do the work.