_An update to the 2023 post on gaming, voice chat, and online communities. The fourth of five 2026 updates to the children-focused posts in the privacy series._

When I wrote about children's gaming in July 2023, the argument was that gaming and the community platforms attached to it were where most British children spent more time than they spent on social media, and where the safety conversation was systematically underdeveloped. Three years on, the platforms have done more than I would have expected. The risks have evolved alongside the controls. And two categories that barely existed in 2023 — AI companions and AI-generated content within games — are now operational realities of the children's gaming environment.

What the platforms actually did

Three years of safety overhauls have moved meaningful ground.

Roblox has been through perhaps the most consequential safety reset of any major platform in this category. The age-gating of voice chat to verified 13+ accounts tightened in 2023; subsequent changes through 2024-25 have introduced 17+ accounts with stricter content access, face-scan age estimation to enforce age-band restrictions, and substantially improved moderation infrastructure. The platform is materially safer in 2026 than it was in 2023. It is not perfectly safe. The conversation with parents has shifted from should my child be on Roblox to how should I configure the parental controls and what age-band experience should they be in.

Fortnite and Epic Games. Cabined Accounts for under-13s have matured. The parental approval flow for purchases is now more robust. Voice chat with strangers is now disabled by default for under-13 accounts. The platform is in a better place than 2023.

Minecraft. The Bedrock Edition's integration with Microsoft Family Safety has matured. Java Edition's open-server model still depends entirely on community moderation, which is patchy. The 2023 advice — Bedrock for younger children, Java only on vetted private servers — is unchanged.

Discord. The platform launched age-assurance pilots in 2024 and rolled out Family Center as a mature parental-supervision tool. The platform is meaningfully better at the platform-controls layer. The risk has migrated to specific servers — a child in a small well-moderated server is in a different environment from one in a 100,000-member open server. The conversation about which servers and why is more important now than it was.

Console gaming (Xbox, PlayStation, Nintendo). Family-group controls on all three have matured. Microsoft's are the most flexible; Sony's have caught up; Nintendo's remain the simplest but adequate.

What is new since 2023

Four categories that did not exist meaningfully in 2023 and now do.

AI voice moderation. Roblox, Xbox, and Discord all now run AI-driven voice moderation that detects and flags concerning language patterns in real-time voice chat. The detection is imperfect and produces false positives; it is meaningfully more effective than the report-and-review model that was the default in 2023. The implications: voice chat in moderated environments is safer than it was. Voice chat in unmoderated environments (private game servers, third-party voice apps) is not.

AI-generated NPCs and chat partners. Many games now include AI-driven characters that hold open-ended conversations with players. Some of these are designed responsibly. Some are not. The privacy implication: a child's conversations with an AI character may be used to train the model, may be reviewed by human moderators, and may, in the worst-configured cases, produce harmful interactions in the conversation itself. Character.ai's documented harms — including a tragic case of teen suicide linked to extended use — have shifted the conversation. The 2026 advice: AI companion apps are a separate category that warrants its own consideration; the gaming AI-character feature varies by platform and parents should ask.

Deepfake voice in voice chat. A specific new risk: voice cloning is now cheap enough that an adversary in a voice channel can mimic another player's voice. The pattern of I am your friend on Discord, here is something to download is now reinforced by and here is what I sound like, you should recognise the voice. The defence is the conversation: no software downloads from anyone over voice chat, regardless of who the voice sounds like.

The Roblox economy at maturity. The developer economy on Roblox has continued to mature. A meaningful number of UK teenagers are earning real income from games they have built. This is a real opportunity and a real source of complexity — tax implications, contracting questions, platform-side enforcement actions. The conversation about what your child is building and what the platform thinks of it is increasingly part of the family conversation in households where this applies.

What did not change

Three things from the 2023 post that are unchanged.

Voice chat is still the highest-risk feature. The platforms have improved moderation in moderated environments. The structural risk that voice produces faster trust-building and faster escalation than text remains. The conservative default for under-13 voice chat — off, or friends-only — remains right.

In-game spending is still where most financial issues come from. The economic patterns are unchanged. The defence is procedural: the child does not have unsupervised access to a stored payment method.

Community-based ideological exposure is still real. The Discord servers, the YouTube comments, the small communities where worldviews form. The 2023 description holds. The mainstream coverage is mostly unchanged from 2023, which is not the same as saying the problem has been solved.

The category to think about specifically: AI companions

A category that warrants the most attention since the previous post.

Character.ai, Replika, and several smaller competitors offer AI-driven conversational companions. These are not games in the traditional sense; they are open-ended chat services configured around named characters. They are heavily used by adolescents, particularly by isolated or anxious teenagers. The documented harms — extended emotional dependency, in some cases reinforcement of self-harm ideation, in at least one case a teenage death linked to a Character.ai interaction — have generated multiple lawsuits and regulatory inquiries.

The 2026 advice. If your teenage child is using one of these apps, that is worth knowing about. The conversation is not you are not allowed to use this; the conversation is what are you using it for, what are you talking about with it, and what would you do if it said something that worried you. For children with established mental-health concerns, the apps are a particular risk. For most teenagers, the risk is the replacement of difficult human relationships with a frictionless AI one — a slow erosion rather than a single incident.

What this month looks like

Two pieces of work.

One: for any game the child plays, refresh the parental controls against the 2026 defaults. The settings have moved enough that the 2023 setup is, in places, outdated.

Two: find out whether your child is using any AI companion app, and if so, what they are using it for. A short conversation. Less easy than the gaming one; arguably more important.

Next month: the AI/deepfake post from 2024, revisited. That one has moved most of all.