For most of the past decade, governments have debated what to regulate about social media: misinformation, pornography, political advertising, hate speech, privacy, bullying and extremist content.
The underlying assumption was that the central problem lay in what appeared on platforms.
The settlement reached with Meta this week suggests that assumption is changing. It marks a move from regulating content to regulating platform design: not only what users see, but how applications are built to keep them watching, returning and engaging.
Meta has agreed to pay as much as $16.7 billion to settle claims by American states that Facebook and Instagram were designed in ways that harmed children while misleading the public about those risks.
The agreement, which remains subject to court approval, follows a federal trial involving a coalition of state attorneys general.
The payment has dominated headlines, but the more consequential provisions concern Meta’s products.
Teenagers will face daily usage limits, overnight access restrictions unless parents intervene, disabled notifications during parts of the school day, hidden like counts and limits on cosmetic-surgery filters.
Meta must also strengthen age-assurance measures and offer a non-personalised feed.
These are not conventional content rules. They do not identify particular posts, videos or opinions for removal. They regulate the design of the platform itself.
Regulating a particular post or video targets content. Limiting how long a teenager can use an application, disabling notifications or changing recommendations targets the architecture through which that content is delivered.
The settlement therefore establishes a new regulatory target: the systems that shape users’ behaviour before any individual piece of content is considered.
Social-media companies have spent nearly two decades building systems designed to maximise engagement.
Infinite feeds remove stopping points; recommendation systems select material likely to sustain attention; notifications encourage users to return; likes and follower counts provide continuous feedback; autoplay eliminates the decision to continue.
None of these mechanisms requires harmful content.
The architecture itself is designed to produce continued behaviour.
The settlement reaches directly into that machinery. A non-personalised feed matters because recommendation algorithms convert observations — what users pause on, skip, share, search for and revisit — into predictions about what will keep them engaged.
That feedback loop is the core of the modern social-media product, and American authorities are beginning to regulate it.
They are doing so without Congress passing a comprehensive national law. State attorneys general have instead used litigation, consumer-protection law and a negotiated settlement to impose detailed product requirements on one of the world’s largest technology companies.
There is an obvious historical precedent. In 1998, America’s major tobacco companies entered the Master Settlement
Agreement with 46 state attorneys general, imposing enormous payments and restrictions on advertising and marketing when federal regulation had proved politically difficult.
Cigarettes and social networks are fundamentally different products, but the institutional mechanism is similar: litigation became a means of changing an industry’s operating rules.
Meta’s agreement is designed not only to change Facebook and Instagram but also to pressure competitors towards similar standards. Part of the payment structure is tied to whether YouTube and TikTok adopt comparable youth protections and make corresponding payments.
The settlement therefore resembles an attempt to establish an industry norm: certain design practices should no longer be acceptable when the user is a child.
Until recently, regulators largely treated social networks as containers for posts, videos, photographs and short clips.
Modern platforms are behavioural systems.
They observe what users watch, ignore, share and revisit, then reorganise the environment around those signals.
The product is not simply the content. It is the interaction between a user and a machine optimised to predict what will hold that user’s attention.
The regulatory question consequently becomes not only what children should encounter, but whether companies should be allowed to deploy particular behavioural mechanisms against them.
Australia has taken a different approach.
Since December 2025, major platforms have been required to take reasonable steps to prevent Australians under 16 from maintaining accounts.
The restriction is not absolute: some under-16 accounts remain, platforms must detect attempts to circumvent age checks, and young people can still access publicly available material without an account.
Australia has primarily regulated access.
America is beginning to regulate architecture.
Australia asks whether a child should be inside the system. The Meta settlement asks what the system should be permitted to do once the child is there.
An age threshold creates a relatively clear boundary. Regulating design requires judgements about the mechanics of persuasion itself.
How long should someone be encouraged to remain online?
When should a platform stop trying to bring them back?
Should engagement metrics be visible?
Should beauty filters be available to children?
When does a recommendation system become an unacceptable instrument of behavioural manipulation?
There are no simple answers. But the distinction between content and distribution is becoming impossible to sustain.
Two teenagers can open the same application and encounter entirely different environments, generated through behavioural signals and an optimisation system designed to predict what will hold their attention.
Regulating individual pieces of content therefore addresses only part of the system.
The deeper power lies in deciding what gets amplified, when it appears, how often it appears and how difficult it is to stop consuming it.
The Meta settlement is an early attempt to place legal boundaries around that infrastructure.
It may also create an uncomfortable question for regulators.
Children are unusually vulnerable to persuasive technologies and deserve additional protection. But the mechanisms being restricted for teenagers do not disappear when users turn 18.
Infinite scroll remains infinite.
Notifications still interrupt.
Recommendation engines continue learning which emotional stimulus will produce another minute of attention.
If these mechanisms are powerful enough to require legal limits for a 17-year-old, what becomes harmless on their eighteenth birthday?
That does not mean governments should impose time limits on adults’ Instagram accounts.
It does mean the intellectual foundations of social-media regulation are shifting.
The first generation of internet regulation concentrated on information.
The next may concentrate on behaviour.
For years, the central political question was what platforms should be allowed to publish.
The Meta settlement points towards another:
What should they be allowed to engineer us to do?