Ask a parent to name social media and you will get the ban list: TikTok, Instagram, Snapchat, YouTube, Facebook, X. Ask a police officer in a child exploitation unit where the grooming conversations actually start, and you will hear a different list — one that begins with a platform most parents file under "games".
This is the definitional problem at the heart of the government's plans, and it matters more than the curfew, the defaults, or any of the machinery I covered in part one. A law that regulates the wrong nouns protects nobody. So before this series gets into federation and onion routing, we need to answer the question the legislation mostly assumes: what is social media?
The definition the law uses
The under-16 ban, following the Australian model, targets platforms whose purpose is open social interaction — public posting, feeds, algorithmic amplification. That test captures the household names. It deliberately excludes messaging services like WhatsApp and Signal, on the architectural argument that closed groups without recommender feeds are a different kind of thing. And it leaves gaming platforms in a promised third category — not banned, but subject to "function restrictions" on things like livestreaming and stranger communication, with the details still to come.
As a piece of drafting, you can see the logic. As a map of where children are harmed, it is upside down.
Where the harm actually lives
Here is the uncomfortable reframe: from a predator's point of view, the feed is not the product. The feed is the shop window. The product is the chat — the private channel where an adult can reach a child, build trust over weeks, and then move the conversation somewhere darker. And chat does not live inside "social media" as the law defines it. It lives inside everything.
It lives inside Roblox — nominally a game, functionally one of the largest social platforms on Earth, with well over a hundred million daily users and an enormous share of them children. It lives in Discord servers, in-game voice channels, Minecraft realms, Fortnite parties, Twitch chats, and the comment sections of platforms whose main product is video. I wrote last year about voice chat and gaming communities precisely because that is where the safeguarding conversations parents aren't having need to happen.
In security terms: the legislation is regulating by brand, but the threat model operates by function. Predators do not follow logos. They follow children, and children are in the chat.
The Roblox object lesson
If you want one case study for why "it's a game, not social media" is a category error, Roblox has spent the last year providing it.
In August 2025, Roblox permanently banned a YouTuber known as Schlep — a young man who, having been groomed on the platform himself as a child, spent years running sting operations against predators using it. Decoy accounts, documented conversations, evidence handed to police; his work contributed to a series of arrests. Roblox banned him under a new policy against "vigilante" investigations, sent a cease-and-desist, and argued — with a straight face — that impersonating minors to catch predators normalises the very conduct being investigated.
There is a legitimate kernel in Roblox's position: freelance stings can genuinely contaminate evidence and collide with live police operations, and no platform can run on vigilante justice. But the optics were catastrophic for a reason that goes beyond optics: a platform that had struggled for years to keep predators out had found the energy to remove someone taking them out. The backlash was enormous. By autumn, the attorneys general of Louisiana and Kentucky had filed suit over child safety failures on the platform.
And then the genuinely instructive thing happened. In January 2026, Roblox began requiring facial age estimation for every user, worldwide, before they can use chat. Not a self-declared birthday — a camera check, applied to the specific function where the harm concentrates.
Sit with the irony. The platform that sits outside the government's definition of social media now runs stronger age enforcement, on the precise function that matters, than anything the banned platforms have been asked to do — and it got there not through the Online Safety Act or the coming ban, but through lawsuits, attorneys general, and a public that watched the Schlep affair unfold. The courts moved faster than the legislation, and they aimed at the chat, not the brand.
That is either an encouraging precedent or a damning comparison, depending on which side of Westminster you sit.
The functional definition parents actually need
So let me offer the definition the legislation should have led with, and the one I suggest you run your household on:
Social media is any product in which a stranger can start a conversation with your child.
That's it. Feeds, follower counts and algorithms are how platforms make money; the conversation is how children come to harm. Apply the functional test and the map redraws itself. Roblox is social media. Discord is social media. The voice channel in a shooting game is social media. The multiplayer chat in a blocky building game rated 7+ is social media. Meanwhile a WhatsApp group containing your child's actual football team is — by this test — relatively low on the list, which is roughly the opposite of how the brand-based instinct ranks them.
The law will not say this, because a functional definition would drag half the games industry into scope and the drafting battle would last a decade. Fine. But nothing stops you using the honest definition at home, tonight.
Auditing by function, not brand
Here is the practical version — one evening, one cup of tea, your child's devices on the table with them present, because this works better as archaeology done together than as a raid:
For every app and game they use, ask four questions. Can strangers contact them here? Not "is it a social network" — can an unknown account send a message, a friend request, a party invite. Is there a private channel? DMs, whispers, party chat, voice — the places moderation cannot easily see. What do the settings currently say? Nearly every platform, Roblox included, has contact controls — friends-only chat, disabled DMs, restricted servers — that default to more open than you would choose. Close them. Where would the conversation move to? Grooming almost never stays on one platform; the move to "let's talk on [somewhere else]" is the single most important tripwire your child can learn to recognise and tell you about — without fear of losing the game, which is the threat predators rely on.
None of this requires Parliament. All of it targets the function the ban list misses.
Where this leaves the law
I said in part one that I would take a blanket under-18 rule, enforced at the device layer, over a curfew with an off-switch. This piece is the other half of that argument: whatever rule we choose only means something if it attaches to the function — stranger contact with children — rather than to a list of brands frozen at the moment of drafting. The government has promised its gaming-platform measures in a later instalment; on the evidence so far, the harms concentrated exactly there are being handled as an afterthought to a ban aimed at the shop windows.
The one hopeful sign in this whole story is Roblox's camera check — proof that age enforcement on the chat function, at global scale, is technically possible when a platform is sufficiently motivated. What motivated them was not a definition of social media. It was consequences.
Next in the series: the federation problem — what happens when the platform is Mastodon, and there is no company to write to.