The conversation about children and technology has placed the burden almost entirely on families. The systems were designed by adults who are not being asked to answer for them

Almost all advice given to parents about children and technology asks something of the parent. Set limits. Monitor use. Have conversations. Model good habits. Keep devices out of bedrooms. Know the platforms.

This advice is sound, and families should follow it. But notice its structure. Every item places the responsibility on the household, and none of it asks anything of the people who built the systems the household is defending against.

That distribution is strange when stated plainly. Parents are being asked to counteract, through vigilance and conversation, the effects of systems designed by well-resourced teams whose professional objective is to hold their child’s attention. A parent working two jobs is expected to out-manoeuvre continuously refined optimisation, in the evenings, with love as their only instrument.

Love is a great deal. It is not a substitute for the systems being built differently.

What Optimisation Does to a Child

The mechanism deserves stating without alarm, because it is not mysterious.

Recommendation systems are optimised against measurable objectives: time spent, sessions initiated, content completed, return visits. The system tests what holds a particular user and delivers more of it. It has no view about whether the content is good for that user. It has a view about whether it works.

Applied to adults, this produces the familiar experience of losing an evening to something one did not intend to watch. Applied to children, it produces something more serious, for reasons that are developmental rather than moral.

A child’s capacity for self-regulation is not fully formed. That is not a character flaw, it is neurology, and it is precisely why childhood is a protected category in every other domain: alcohol, gambling, employment, contracts, consent. We do not expect children to exercise adult judgement against adult pressures, and we do not blame them when they cannot.

Yet in this domain we have effectively asked exactly that, and then framed the resulting harm as a parenting failure.

The specific patterns are documented well enough to name. Systems that learn a child responds to content about appearance and supply escalating quantities of it. Feeds that discover distress content holds attention and deliver more to the young person already struggling. Design features borrowed directly from gambling: variable rewards, streaks that punish absence, infinite scroll with no natural stopping point. Social metrics that render a child’s standing among their peers as a public number, updated continuously.

None of these were built to harm children. They were built to work, and they work.

The Shape of the Failure

There is a further problem in how these systems handle children, and it is structural rather than incidental.

Age verification is largely fictional. A child enters a date of birth and is admitted. Platforms know this. The knowledge does not create urgency, because a rigorous age gate would reduce the user base, and no commercial incentive exists to reduce the user base.

Default settings favour exposure. Accounts are frequently public unless changed, discoverable by strangers unless changed, and set to maximum data collection unless changed. Every default is a decision, and defaults that require a knowledgeable adult to correct are decisions about which children will be protected.

Reporting mechanisms are built for scale rather than for children. A young person who encounters something distressing, or who is being contacted by an adult, faces a form designed for volume processing. The response, if it comes, arrives days later from no one in particular.

And harm is externalised almost completely. When a platform’s design contributes to a child’s distress, the cost is borne by the child, the family, the school, and the health system. It does not appear on the platform’s accounts. This is a textbook externality, and externalities do not resolve themselves.

Where Responsibility Actually Sits

Catholic social teaching offers a framework here that is more demanding than it first appears.

Human dignity is not conditional on age, and a child’s dignity is not less because they are not yet a market participant with negotiating power. A system that treats a twelve-year-old as an engagement opportunity has already made an anthropological claim, and it is the wrong one.

The preferential option for the vulnerable identifies who a system should be judged by. Not the median user, not the sophisticated user, but the one least equipped to protect themselves. By that measure, a platform performing acceptably for adults while harming children has failed, whatever its aggregate metrics show.

Subsidiarity is frequently invoked to argue that families should decide these matters. It says something more balanced. Decisions belong at the level closest to those affected, and higher levels must supply what that level cannot provide. A family can set rules in its own home. It cannot audit a recommendation algorithm, compel a design change, or verify what a company does with its child’s data. Those things require authority the family does not have, which is precisely when subsidiarity requires the higher level to act.

Solidarity notes that the children most exposed are the ones with the least protection: whose parents work long hours, who lack a supervising adult after school, who received a device without accompanying formation. A regime that protects only the children of attentive, technically literate households has protected the already-advantaged.

What Should Be Required

Not vague commitments. Specific obligations.

Safety as the default, not the setting. Accounts belonging to minors private by default, discovery by strangers off by default, data collection minimised by default. If a protection requires a knowledgeable adult to activate, it will not reach the children who most need it.

No optimisation for engagement on minors’ accounts. This is the central one. A recommendation system serving a child should not be optimised for time spent. Chronological or explicitly non-engagement-based feeds for minors are technically trivial and commercially unattractive, which is exactly why they require regulation rather than encouragement.

Design features borrowed from gambling prohibited for minors. Streaks, variable rewards, infinite scroll, and artificial scarcity are known persuasive mechanisms. They are regulated in gambling because they work. They work on children too.

Age verification that functions. Meaningful, privacy-preserving, and enforced. The technical objections are real and they are not insurmountable.

Independent audit of systems serving children. By bodies with access to the actual algorithms, not to summaries prepared for publication.

Liability for foreseeable harm. Until the cost of harming children appears on a balance sheet, it will remain someone else’s problem. Every other industry that endangers children has learned this through liability rather than through persuasion.

Not Instead of Parents

None of this diminishes what families do. Formation still happens at home, and it is irreplaceable. A child who has been taught to recognise manipulation, who has adults who listen without interrogating, who has experiences of attention and boredom and unmediated friendship, is more resilient than one who has not. That work remains essential.

But it should not be the whole defence. Parents should not have to be the sole barrier between their children and systems built by people who never have to look at the result.

We do not ask parents to test food for contamination, verify the crash performance of a car seat, or check a medicine for adulteration. We require the producers to make products that do not harm children, and we hold them liable when they do. We do this because we recognised, eventually, that individual vigilance does not scale and that placing the entire burden on parents means the least-resourced children pay.

The same recognition has not yet arrived in this domain. Until it does, families will keep doing what they can, and children will keep encountering systems that were optimised against them by people who will never meet them.

That is not a technological problem. It is a decision, and decisions can be made differently.