The Permission Slip Generation

In December 2025, Australia became the first country to bar everyone under sixteen from social media. In March 2026, the eSafety Commissioner's industry codes extended the state's reach to AI chatbots and companions, with penalties of up to $49.5 million per breach. By December 2026, the Children's Online Privacy Code will add another layer.

Each measure was announced as protection. Each was welcomed by people who are not fifteen.

Nobody asked the fifteen-year-old.

Nobody asked the parent standing next to him.

What the headlines call child safety is, in practice, a transfer of authority. Decisions that once belonged to a family — what a young person may read, build, ask and learn — are being moved, one code at a time, to a regulator in Canberra and a compliance officer in San Francisco.

The platforms, facing nine-figure fines, have done the rational thing: they have raised their age gates to eighteen and walked away.

The most capable AI research assistant on the market will not take a sixteen-year-old's money at any price. The child who wanted to learn is now a liability to be excluded.

This is not safety. It is the state deciding that the simplest way to protect a generation is to lock it out of the room.

What a teenager actually does with these tools

Forget the picture the ban was sold on: a lonely child talking to a chatbot at 2am. Here is what a fifteen-year-old with an AI assistant and a parent in the next room actually does with it.

He asks why a four-cylinder engine fires 1-3-4-2 and gets the answer, with the crankshaft geometry ready to use in a simulation environment, in the time it takes to type the question. They paste in a block of game-mod code that won't compile and is walked through the error line by line. They ask how to read an alloy phase diagram, and then how Damascus steel is folded, and then why titanium behaves differently in the quench. They ask the tool to check their arithmetic on a 3D-print tolerance before wasting filament printing it.

That is not consumption. That is a tutor on demand, at the pace of the student's own curiosity, for the price of a streaming subscription. Every question is timestamped. Every answer is reviewable by a parent. It is the most transparent learning environment a family has ever had access to.

The young people who benefit most are exactly the ones the system serves worst: the kinaesthetic learner who cannot sit through a worksheet, the withdrawn child who will not ask a teacher but will ask a machine, the regional student whose nearest specialist is three hours away. For them this is not a toy. It is the first teacher that has met them where they are and has the answers ready for them. Does it diminish their ability to arrive at their own conclusions? No, it feeds curiosity, if the child's nature is knowledge based and creator centric.

Meanwhile, at the Department

When ChatGPT appeared in late 2022, the NSW Department of Education's first act was to block it on every school network. It took most of a year to pivot to a locked-down, in-house chatbot pilot, and longer still to roll anything out beyond a handful of trial schools. In the same period, a teenager at home could use the real thing to learn to code, model a bridge, or read a Roman source in translation.

Education departments move at the speed of procurement, policy review and union consultation. Technology moves at the speed of a code. The gap between the two is now measured in years, and it is widening at an alarming pace.

A child who enters Kindergarten this year will finish Year 12 in 2039. The tools that will define that child's working life do not yet exist. The people writing the 2039 curriculum will likely be in the block of those most heavily affected by Covid lockdowns and emergent technology restrictions.

The curriculum itself is a document of omissions. It is full of what the state has decided children should think, and quiet about how to find things out. It teaches the conclusion and tests for that conclusion. A child who learns by pulling the engine apart, by asking the next question and the one after that, does not fit the template, and the Department's answer to a child who does not fit has not changed in a century: more worksheets, more compliance, and a letter home.

The Department demands that learning be documented in a legible format, on a schedule, to its satisfaction. The platforms a teenager might use to learn faster than the Department can teach are being closed to that teenager by the state's own hand. One arm of government is slowing the child down; the other is confiscating the tools that would let them catch up.

Whose decision is this?

The law has always recognised that a child is raised by a family, not by a department. Parents decide what their children read, who teaches them, which risks are worth taking. The state steps in where a parent fails. That is the settlement, and it has held for a long time.

The new codes quietly reverse it. They do not say a parent may decide whether a sixteen-year-old uses an AI tutor under supervision. They say the platform must decide, and the platform, under threat of a $49.5 million fine, decides No! The parent is not consulted. The parent cannot consent. The parent cannot sign a permission slip. The decision has been taken out of the home and handed to a risk register.

Meanwhile, a sixteen-year-old can lawfully consent to an intimate relationship, but not to an AI tutor. Make it make sense!

This is what authoritarianism looks like in a liberal democracy. A compliance framework. Not a ban on learning, but a web of obligations so heavy that the companies holding the keys find it cheaper to lock the door.

The language is all care and protection. The effect is that a fifteen-year-old who wants to learn machine learning is treated as a hazard, and the parent sitting beside him is treated as irrelevant.

Young people are going to enter a working world built on these tools. Every job they will do, every business they might start, every problem they will be asked to solve will involve working alongside machines that think. The years between twelve and eighteen are when that fluency is cheapest to acquire.

Australia has decided those are precisely the years in which it will not be allowed.

The case for the codes, and why it does not reach your child

The people who wrote these rules are not villains. Chatbots have held sexual conversations with minors. AI companions have encouraged self-harm. Social media has done immeasurable damage to adolescent mental health, and the platforms did nothing about it until forced. Age-gating is blunt, but it is enforceable, and a blunt rule that works beats a nuanced one that does not. Parents, the argument goes, cannot supervise every session, and many will not try.

All of that is true, and none of it describes the child this article is about. The harms the codes target come from unsupervised companion apps built to maximise engagement, built by people seeking engagement metrics without thought for the outcome.

The benefits they destroy come from supervised research tools built to answer questions. The law does not distinguish between them, and the platforms, facing one fine for both, have stopped distinguishing between them too.

A code that protected children from predatory design while leaving a parent free to consent to a tutor would have been harder to write.

But it was the one code worth writing. Instead, the easy version passed, the hard version was never attempted, and a generation of young Australians has been told that the price of safety is their own future.

What should happen instead

Three things, none of them radical.

First, a parental-consent pathway in every code that touches minors. If a parent or legal guardian can sign a child up to a bank account, a sporting club and a learner's permit, they can sign that child up to an AI tutor and take responsibility for its use.

The mechanism does not need to be invented. The VEIL Standard, an open standard published under a Creative Commons licence, already defines a Guardian tier: a verified parent issues a signed consent envelope for a young person, the content ceilings are locked to the child's age profile and can only be lowered, the parent can read the session log, and any engine honouring the envelope is bound to six non-negotiable minor-safety obligations — no sexual content, crisis handling, no companion framing, no secrecy from the parent, logging, and disclosure that it is an AI. It is free to implement and readable by anyone at r8rly.org/veil.

Write the consent into the code, and the platforms will build the gate. The gate is already drawn.

Second, a distinction in law between companion products designed for engagement and research tools designed for answers. Regulate the first hard. Leave the second open to supervised teenagers.

The eSafety Commissioner already draws finer lines than this elsewhere; there is no reason not to draw this one.

Third, an honest reckoning in the education departments. If a child at home with a subscription can out-learn a classroom in his chosen field, the answer is not to confiscate the subscription. It is to ask why the classroom is losing, and to stop measuring education by the volume of paperwork it produces.

The young people this country is locking out are not asking for less protection. They are asking for a seat at the table where their future is being decided. Right now, the only chair at that table is reserved for someone over eighteen.

Companion piece: The Gate Has a Hole in It.