One caveat before we begin. This bill is moving through a live political moment, and the vote counts and timings below — who will back it, and whether it is voted on before the election — can change within a day of publication. Take them as accurate on 1 September 2026, and no further. The argument does not depend on the day’s arithmetic.

On 24 August, Christopher Luxon unveiled a bill to bar New Zealanders under sixteen from social media. It arrived with the usual furniture of a child-protection announcement: a grave line about harm being done to a generation, a statistic about screen time, a list of the platforms every parent already fears. What it did not arrive with was the numbers to pass on its own. Two of the three parties in Luxon’s own government, ACT and New Zealand First, refuse to vote for it. That should have been the end of it. Instead the opposition rescued it: Labour has offered National the votes to carry the bill through a first reading, and urged the government to hold that reading before the election. And even so it may not happen, because the government now says there is not enough time to bring it to a vote before Parliament rises for the 7 November election.

Let your thoughts rest on that before anything else. A government that could not carry its own coalition was handed the votes by its rivals, and is letting the clock run out rather than use them. It found time for the announcement. It says it cannot find time for the vote. That tells you what kind of object this is. It is a law meant to be seen more than it is meant to be enforced.

A law built to be seen, not to work.

I want to be fair to the impulse behind it, because the worry is real. Adolescent anxiety and depression have climbed alongside the smartphone, and “we can’t yet prove causation” is a thin reason to do nothing when the subject is a developing brain. A government that acts early and imperfectly may age better than one that waited for certainty. So the goal is not the problem. The instrument is.

What Australia already shows

Consider the evidence we actually have. Australia’s world-first ban took effect last December, which makes it the nearest thing to a controlled test anywhere in the world. A peer-reviewed study in the British Medical Journal, published in June, found little sign that Australian teenagers had left social media as a result: three months in, 85 per cent of those aged twelve to fifteen were still using it. Set that beside the other Australian number now doing the rounds: in the month after the ban, platforms told the regulator they had revoked around 4.7 million accounts belonging to children.

Both figures are true at once, and the space between them is revealing. Four-point-seven million accounts closed is an output. Children spending less time in harm’s way is an outcome. The bill, the coverage, and the podium all reach for the output, because the output is easy to produce and easy to report. The child whose account is shut opens another, borrows a sibling’s login, routes around the block with a VPN, or moves to a platform the ban does not touch. The regulator gets a clean, enormous figure. The problem the figure was meant to describe does not move. If there is one lesson to take from this, it is that a law can generate a spectacular statistic and change nothing it set out to change, and the statistic is the only part that survives.

Checking everyone to catch a few

Then there is the machinery, and here the bill is better than its critics, mine included, tend to assume. To keep under-sixteens off a platform you have to work out how old everyone is. The draft is careful about how. It forbids relying on a passport or a digital ID alone, throws out the useless date-of-birth box, and pushes instead toward age estimation — a scan of your face or your behaviour that guesses whether you are old enough. It orders that whatever is gathered for the check be used for nothing else and destroyed once the check is done, with protections beyond the Privacy Act. Someone saw the surveillance checkpoint coming and drafted against it. Credit where it is due.

A lone figure faces a long colonnade of tall translucent amber panels, each holding the shadowed silhouette of a person, a thread of light running across them — a hall of screened figures.

But grant the safeguards their best case and look at what still stands. To find the under-sixteens you must screen everyone: every adult’s face or behaviour run through an age-estimation system, because you cannot check the child without checking the crowd he stands in. That estimation is least reliable at exactly the fifteen-to-seventeen line that matters, so the friction and the false readings fall on adults too. And a rule that the data be destroyed is a promise, not a property of the system. It does nothing against a breach while the data still sits there, and nothing against a court order or a police request for it. The bill does not build a surveillance state. It builds a surveillance moment, over the whole population, and trusts that the moment closes cleanly every single time. Backed by a fine of up to forty million dollars or a tenth of global revenue, that is a great deal riding on trust — and trust is not a safeguard.

The one exemption that doesn’t fit

The bill also declines to treat the whole internet as a single object. It gates the high-risk feeds — Instagram, TikTok, Snapchat, and Facebook — while leaving messaging alone, and it reaches past the obvious targets to pull AI companion apps inside the net, which is more foresight than most of these laws manage. Someone thought about where the harm actually lives. Which makes the one exemption that does not fit harder to explain, not easier.

Roblox is waved through as a game. Set that against the record. A 2024 Bloomberg Businessweek investigation documented at least two dozen arrests in the United States since 2018 of adults who abused or abducted children they had met or groomed on the platform. Among them were a sheriff’s deputy, a third-grade teacher, and a nurse: not names from the margins, but people trusted with children. Since then multiple state attorneys general have moved on the company, and Florida’s went as far as criminal subpoenas.

Now read the bill against that record. It would force Instagram to verify every user’s age or pay up to a tenth of its global revenue in fines, because a teenager’s feed is judged too dangerous to leave unguarded. And it exempts the platform where the danger is not a diffuse worry about mood and sleep but a column of arrest records. The harm on Instagram is real and hard to measure. The harm on Roblox is specific, criminal, and already written down. The bill has them the wrong way round.

The good law inside

And yet there is a genuinely good law inside this one. Its full name is the Online Safety (Minimum Age and Child Safety Risk Assessment) Bill, and that second clause is valuable. Platforms used by children would have to assess and report on the risks they create. That is the right lever. It goes at the mechanism, the design, the algorithm, and the defaults, rather than at the user’s birthday. It does not require an identity checkpoint over everyone. It does not depend on stopping a determined teenager, which no law has ever managed. The pity of the bill is that it carries its own better answer and then buries it beneath the half that will fail.

The same week, a working lever

You do not have to guess whether that lever works, because it was pulled the same week. While New Zealand tabled a ban its own government would not find time to put to a vote, forty-seven American states settled with Meta for up to eighteen billion dollars and made it do what the risk-assessment duty only gestures at. Meta must now switch its teen-safety settings on by default — a non-personalised feed, hidden like counts, and capped screen time, each locked unless a supervising parent unlocks it — reversing the opt-in model it defended for years. The lever that works arrived through a courtroom. The one that does not is stuck in a parliament.

And here is the number that should end the argument. A Colorado prosecutor put teen uptake of Take a Break — the screen-time tool Meta has paraded for years as proof of its good faith — at 1.8 per cent, and Instagram’s chief conceded the opt-in rates are low. The feature was never missing. It existed, and almost nobody touched it, because opt-in keeps the safety notional and the take-up negligible. Turn it on by default and you change the experience of every teenager on the platform, not the tiny fraction who went looking for it. That is the whole difference between a safety feature that exists and one that is used, measured in a single figure. The settlement is not a clean win. To tell teens from adults it leans on the same kind of age-screening the New Zealand bill would, and the same privacy cost rides along, which is the case for building the age signal into the device rather than into every app. But on the thing that counts, it moves the default for everyone, not the few who chose it.

The platforms it cannot reach

That contrast has a sting the bill cannot pull out. America reached Meta because Meta answers to American courts: a domestic defendant, domestic assets, a judgment that bites. New Zealand points at the same giants from the far side of the world. Its bill reaches across borders on paper, but the regulator would be chasing companies with no presence here, no duty to answer local process, and no local assets to seize. The lawyers reading the draft say the consequence plainly: the regime bites the small and the locally present, while the largest platforms — the ones that matter most — stay out of reach unless they choose to cooperate. The fine that looks so fierce falls hardest on those least able to dodge it and glances off those most able. New Zealand can order its own providers and app stores to block a platform, a real power and a blunt one. What it cannot do is what America just did: make the company itself change.

The children the ban forgets

A small child stands at the edge of a vast dark maze, a single glowing orange path winding through it toward a lit gateway.

There is one more cost the protective framing steps around. Children have rights to take part and to speak, not only to be shielded. For the isolated child — the rural one, the one in an unsafe house — the platform is often the line out rather than the threat. A blunt ban cuts the line and the threat together. This is not sentiment. It is the objection ACT and New Zealand First are making, and it now has judicial company. On 14 August, France’s Constitutional Council struck down an under-fifteen ban on exactly these grounds. The court held that the measure disproportionately infringed freedom of expression, that it was too broad to weigh either the situation of an individual child or the specific risk of a given platform, and that it raised privacy concerns of its own. Read that back against the bill in front of us. Too broad, blind to the platform, careless of privacy: the French court named the same three faults from the bench that this argument names from the outside. That tension is not a detail to be managed away. It sits at the centre of the question.

Where I land

So here is where I land. The bill is better built than a first glance suggests, and worse justified. The function-based carve-outs and the risk-assessment duty show someone thinking hard. The one incoherent exemption and the age gate show the thinking losing to the politics. Strip the prohibition. Keep and strengthen the risk-assessment duty. Put the age signal, if there must be one, at the level of the device or the operating system, where it costs far less privacy than turning every platform into a border post. None of that photographs well, which is rather the point.

My children are not made safer by a law that looks like it is protecting them. They are made safer by one that does. This bill knows the difference. It chose the other thing.

— · —

Sources

Lex Praxis advises New Zealand employers on employment law and the governance of AI and technology in the workplace. This article is general commentary on proposed legislation and does not constitute legal advice. For advice specific to your circumstances, contact us directly.