For years, the debate over regulating social networks has focused primarily on what platforms should remove: terrorism, abuse, disinformation, illegal content, and material harmful to minors. Australia is trying to shift the target further upstream. Not just the content itself, but the mechanism that decides which content reaches the user’s eyes.

On September 8, Anthony Albanese’s government published draft legislation for a new Digital Duty of Care that includes a measure with an intentionally simple name: “My Feed, My Way”. If the proposal becomes law, social networks will have to offer users over the age of 16 an explicit choice regarding their main feed: personalized content recommended by the algorithm, or posts from friends and creators the user has chosen to follow.

The structure of the proposal is key. It is not simply a matter of hiding an existing toggle deep in the settings. The government wants platforms and new users to receive a clear notification and be presented with the choice of their default feed. In other words, the algorithm should no longer be treated as the natural, inevitable default of the social network, but as a mode that users can accept or reject.

It should be emphasized that the law has not yet been passed. The government released an exposure draft, open for consultation until September 22, and stated its intention to introduce the bill to Parliament by the end of the year. But precisely because it intervenes in platform design and defaults, the Australian project could have consequences reaching far beyond its domestic market.

The point is not to ban the algorithm

“My Feed, My Way” does not eliminate recommendation systems. A user will still be able to choose a personalized feed, discover unknown accounts, receive videos or posts selected based on their activity, and use the social network exactly as they do today. The difference is that this mode must become a visible choice.

The distinction matters because platforms have built an increasing part of their product around recommendation systems. Instagram, TikTok, Facebook, YouTube, and similar services no longer merely show what we post or what the people we follow post. They try to predict which content will keep us around longer, make us react, share, or return to the app.

From a technical standpoint, it is an extremely powerful feature. A recommendation system can surface relevant content among billions of items and can highlight creators that a user might never have encountered. Economically, it is even more critical: the better a platform understands what holds attention, the more it can increase time spent on the service and the value of its advertising inventory.

The Australian government does not question the existence of this model. It questions whether it should be the default.

Regulation shifts from content to architecture

This is the most compelling aspect of the proposal. Traditional content legislation requires platforms to intervene after specific material has appeared. By contrast, a rule on recommendation systems intervenes in how the product distributes and amplifies that material.

The Australian draft introduces a duty of care: digital providers would need to identify foreseeable risks, document the measures taken to mitigate them, and verify over time that those measures remain effective. The government envisions an oversight role for the eSafety Commissioner, with fines that could reach up to 109.2 million Australian dollars for failing to comply with these obligations.

For minors, the scope extends well beyond social media alone. The government explicitly states that the duty of care must also cover chatbots, gaming, apps, and messaging services, with the goal of reducing exposure to features and content that could cause harm. The draft also provides for swifter enforcement powers against “nudify” apps and websites—services that generate non-consensual synthetic intimate imagery.

The overarching rationale is that a digital service should not be assessed solely by what it hosts, but also by the foreseeable consequences of its design choices.

Why the default is a matter of power

In theory, many social networks already offer forms of control over the feed. The problem is that what exists in the settings does not necessarily match what exists in the actual experience.

An option hidden in a menu, requiring periodic reactivation, or made less convenient than the standard mode may be formally available yet practically irrelevant. The Australian government appears intent on intervening precisely on this point: users must choose their primary feed, and platforms must respect that choice.

During a press conference on September 8, Minister for Communications Anika Wells said that Canberra has observed what happened in Europe and wants to avoid forms of “malicious compliance”—meaning compliance that is merely formal and makes the user's right difficult to exercise in practice. The Australian proposal therefore aims for an explicit choice regarding the default feed, which can be revised whenever the user wants to change their mind.

It is a seemingly small difference, but defaults are one of the most powerful levers in digital design. Most people do not alter many initial settings. If the algorithmic feed is turned on by default, the platform starts with an advantage. If users are asked which mode they want, the responsibility of choice shifts.

Europe has already paved the way, but Australia wants to make the choice more evident

The concept does not emerge in a vacuum. The European Digital Services Act requires very large online platforms and search engines using recommender systems to offer at least one option not based on user profiling. In addition, the DSA requires transparency regarding the main parameters of recommender systems.

However, Canberra is attempting to translate that principle into a much more straightforward action for the consumer: do you want a personalized feed, or do you primarily want to see accounts you have chosen to follow?

This simplicity can have significant effects. Digital regulations often fail not because a right does not exist, but because that right is difficult to find, understand, or exercise. Making the choice part of onboarding or the core experience makes user control far more visible.

It is not yet clear how individual platforms would have to technically implement the distinction, how much room would remain for chronological ordering, contextual signals, or non-profiled recommendations, and how services with hybrid products would be treated. It is precisely these details that the consultation phase will need to clarify.

The economic problem: the algorithmic feed is not an accessory

For Big Tech, a non-personalized feed is not simply an alternative interface. It can change user behavior and therefore the business model.

Recommendation systems are built to increase the likelihood that users find something interesting even when the people they follow are not posting. Without that engine, a portion of the platforms risks becoming less “infinite”: less new content, fewer surprises, potentially less time spent in the app.

This does not automatically mean that a non-algorithmic feed will reduce revenue. Platforms have many other forms of personalization and advertising at their disposal. But it does mean that a regulatory choice regarding design can strike directly at the heart of the attention economy.

This is also why the issue quickly took on a geopolitical dimension. The Guardian reported on September 9 that the US administration intends to raise objections to the Australian plan. Since 2025, the White House has maintained a very critical stance toward foreign taxes, fines, and regulations deemed discriminatory or excessively burdensome for American tech companies.

The tension is predictable: many of the platforms that would need to adapt are American. But the legal crux will be understanding whether the Australian rules are applied in a general and proportionate manner to all services that meet the criteria, or whether they could indeed be interpreted as a disproportionate burden on specific foreign companies.

The opposite risk: believing that the chronological feed is automatically “good”

There is also a mistake to avoid in the narrative surrounding the reform: a non-personalized feed is not necessarily healthier, fairer, or more democratic.

The accounts a person chooses to follow can spread extreme, false, or harmful content. A chronological order can favor those who post the most. A social network built over time can itself become a very tight bubble. And recommendation systems can also be used to reduce certain risks, for instance by avoiding showing content already flagged as sensitive.

The Australian measure makes sense primarily when read as a rule on user autonomy, rather than a declaration that the algorithm is always bad.

The eSafety Commissioner itself, in its document on recommender systems, acknowledges that these tools can improve the experience and help people find relevant information. The problem arises when systems optimize for engagement without adequately distinguishing between positive and negative interactions, or when they build up cumulative exposure to harmful content.

The real novelty is demanding product accountability

For a long time, platforms argued they were primarily intermediaries: environments where users post content and interact. But the algorithmic feed makes this description increasingly inadequate. The platform does not merely host: it selects, orders, amplifies, and decides which item to display now and which to leave invisible.

The more decisive this curation is, the harder it becomes to separate the content from the design that distributes it.

The Australian Digital Duty of Care attempts to formalize this insight: if a feature produces foreseeable risks, the company must assess and mitigate them. It is not enough to say that the user clicked, watched, or engaged voluntarily if the entire product is built to predict and influence those very actions.

It is a regulatory model that could extend well beyond social media. AI agents, companion chatbots, online video games, and apps using personalization systems are introducing increasingly adaptive mechanisms. The question will be the same: when software learns to shape its behavior around the user, what obligations do its designers have?

Why this proposal should be followed

The Australian draft could be amended, weakened, or even fail to become law in its current form. The text is open for consultation, and many details remain undecided. It would therefore be premature to describe it as a new global standard.

Yet the direction matters. After years of governments and platforms arguing primarily about moderation, attention is shifting toward the architecture of digital products: recommendation engines, infinite scroll, notifications, defaults, profiling systems, and AI tools that alter the experience in real time.

If “My Feed, My Way” does come into force and is implemented without shortcuts, it could provide a very useful experiment: revealing how many people, when the choice is made genuinely simple, prefer the algorithm and how many prefer a feed built around their own relationships.

It would also be a test of the true power of recommendations. Today, the “For You” feed feels like an inevitable feature of the internet. Australia is proposing to treat it for what it is: a product decision.

And product decisions, when they affect millions of people, can become political decisions.

Sources