Grooming signals
Building trust, isolating a child, moving a chat to a private app, and other patterns that repeat across a conversation.
What we detect
What Lantore looks for, and what it honestly cannot do
Lantore is a smoke detector, not a guarantee. It watches for patterns that matter to child-safety experts, explains what it found, and is upfront about gaps.
Plain language
What Lantore looks for
Lantore is engagement-first: it only reads while your child is taking part in a chat or voice session, and for a few minutes after. Within those conversations it looks for things like:
Someone moving the conversation to a different app or platform
Snapchat, Discord, Instagram, Telegram, WhatsApp, text messages, a private server. It is one of the most common steps in online grooming.
Why it matters: moving a chat to a private app takes it out of view of the game's safety tools and of you.
On its own, a move is often harmless, so it is a quiet, low-priority heads-up in your alert queue, never a push notification. It is on by default for children under 13 and off for 13 and up, and you can change it for each child.
Other warning signs raise its priority. When a move comes together with other signs in the same conversation, you always get a higher-priority alert, whatever the heads-up setting. Those signs include:
- special attention or compliments
- requests for secrecy
- personal, age or location questions
- gifts or in-game currency
- pressure or urgency
- arranging to meet
- requests for photos
Lantore looks at patterns across a conversation, not single words or set phrases, so it will not catch everything and will sometimes raise things that turn out to be fine.
An example alert. Other warning signs in the same conversation raise how urgent it is.
Categories
The patterns we watch for
We look at patterns across a conversation, not single words. A swear word is not an alert; a stranger steering the chat somewhere private, especially alongside other warning signs, is.
Sextortion
Requests for private images followed by threats or pressure to pay or comply.
Sexual content
Sexual talk or requests directed at, or involving, your child.
Bullying and harassment
Repeated put-downs, exclusion, and pile-ons, from more than one person or over time.
Self-harm
Signs that your child may be struggling. Handled gently, privately, and never shared onward.
Violence and threats
Threats against your child, or by your child against others.
Personal information
Your child sharing, or someone asking for, address, school, phone, or when they are home alone.
Scams and currency theft
Free currency offers, fake giveaways, account takeovers, pressure to buy gifts.
Moving off-platform
Pushing the chat to Snapchat, Telegram or another app where you cannot see.
Hate and extremism
Hateful or radicalizing language directed at or around your child.
Drugs and alcohol
Offers, sales or pressure involving substances.
Age and location probing
Asking how old your child is, where they live, or whether they are alone.
Severity
Four levels, each with a plain meaning
FYI
Worth knowing, no action needed. Collected in a daily digest by default.
Worth a conversation
Something to talk about calmly in the next day or so.
Urgent
A concerning pattern. Look soon. Text messages fire from this level.
Emergency
Possible immediate harm. The alert opens with a plain "what to do now" panel and pointers to emergency and reporting help.
Coverage honesty
What we can and cannot see
Lantore is in beta and Windows only. We are testing game by game, because the way a game draws its chat changes with resolution, mode and updates. Here is where things stand.
| App | Chat text | Notes |
|---|---|---|
| Roblox | Beta, testing | Read from the screen. In-experience chat varies by game; we are building a per-game test matrix. |
| Discord (desktop and web) | Beta, testing | Read from the screen. We never use Discord accounts, bots or APIs, and we do not read anything that is not on screen. |
| Fortnite | Beta, testing | Screen reading. Games with anti-cheat that blank screen capture may show as "cannot see" on the Devices page. |
| Minecraft | Beta, testing | Screen reading for Java and Bedrock chat. |
| Other games and chat apps | Best effort | Lantore looks for chat-like text in apps it does not know, but accuracy is lower than for the apps above. |
| Voice chat | Optional | Off unless you turn it on. Speech is turned into text on the PC; audio is never stored or uploaded. There is a clear notice, and laws about recording other people vary, so please read the notice when you enable it. |
| Images, video, voice notes in chat | Not covered | We read text only. |
| Phones, tablets, consoles, Mac | Not covered | Windows 10 and 11 only for now. Other platforms are planned, not available. |
| Languages other than English | Weaker | Text recognition and detection are best in English. Accuracy elsewhere is lower and not yet measured. |
Engagement-first: when Lantore reads
Lantore reads chat and voice only while your child is taking part (typing in a chat or speaking in a voice session) and for a few minutes after. It does not read what other players say when your child is not involved. Typing outside chat, like documents, searches and passwords, is never captured.
When Lantore cannot see
Some games block screen capture (certain anti-cheat or exclusive-fullscreen modes). If Lantore sees only black frames, it reports "cannot see the screen" on your dashboard and can notify you, instead of quietly showing nothing. Switching a game to borderless windowed mode often fixes it.
About accuracy
Detecting harm in fast, slangy game chat is hard. Early research on real game chat shows automated systems produce many false alarms. Lantore is tuned to keep alerts few and explained, and "mark as fine" teaches it. We have not yet published accuracy figures, and we will publish them, with how we measured, once we have solid data rather than claim numbers now.
Gentle with private topics
Words about identity, sexuality or mental health are not signs of danger by themselves, and a school-style "keyword" alert can do real harm by outing a young person. Lantore treats these as private: it flags concerning behavior, not topics; softens the wording; and does not suggest sharing or escalating the alert to anyone else. If something suggests your child may be struggling, the alert points you toward warmth and support first.
If you get an emergency alert
- Stay calm and make sure your child is safe. They are not in trouble.
- If anyone is in immediate danger, call your local emergency number (911 in the US).
- Keep the evidence. Do not reply to the other person or pay anyone.
- In the US, report online enticement and sextortion to the NCMEC CyberTipline (1-800-843-5678). Take It Down can help remove intimate images of a minor.
- If your child may be thinking about harming themselves, call or text 988 (US), or find a local helpline at findahelpline.com.
Lantore never contacts the authorities or anyone else for you. You decide what happens next.
Start with a calmer way to look out for them
Free during the beta. Set up takes about five minutes, and you will do it together with your child.