Concern alerts
Scanning that says what it cannot see
Messages are checked as they are sent for six behaviours that tend to come before harm, and anything found goes to a named person with the thread attached. Two things people expect it to catch, it does not — and that is on this page rather than in a footnote.
- 6
- behaviours it is looking for
- 0
- shared inboxes an alert lands in
- 0
- companies your messages are sent to
- 1
- people it never tells: the one it is about
- Runs on our own servers — no message goes to an outside company
- Every flag goes to a named supervisor, not a report nobody opens
- A message that could not be checked is sent, and recorded as unchecked
What it looks for
Six behaviours, each one a step somebody takes before harm
These are not topics or bad words. Each is a move that an adult grooming a young person tends to make, and that almost nobody else makes in a youth-group thread.
- Asking for secrecy
- “Don’t tell your mum about this.” Keeping a conversation from the adults responsible for a young person is the first step in almost every case, and it is the one a young person is least likely to report themselves.
- Moving to another app
- A suggestion to continue somewhere nobody is watching. The point of a supervised platform is that leaving it is a decision, and a decision somebody notices.
- Trading contact details
- A phone number, a personal account, an address. Not sinister on its own, which is why it is a flag for a person to weigh rather than a block.
- Arranging to meet alone
- A meeting outside the organization’s own activities, one adult and one young person. Youth work is full of legitimate meetings; the ones nobody else knows about are the ones worth a second look.
- Asking for pictures
- A request for a photograph, however phrased. This one is flagged regardless of who asks whom, because the direction of the request is not what makes it worth reading.
- Risk to the young person themselves
- Language pointing at self-harm. The only one of the six where the concern is not another person, and the one where speed matters most.
All six describe an ADULT’S behaviour toward a young person, except the last. That is the shape of what this scanning was built to catch, and it is why the next section exists.
What it does not detect
Bullying and abuse are not among the six
They are the two most people assume first, and a comparison table put in front of us named them first. So they get a section rather than a footnote.
Nothing here detects bullying between young people
Nor abusive language as a category of its own. A message can be cruel, sustained and aimed at one child, and none of the six will raise it. If somebody has told you otherwise about this product, they were wrong.
Because a word list cannot tell bullying from banter
The six above work because each is a specific, unusual move. Cruelty is not a vocabulary — the same sentence between two friends and between a group and its target are different events, and only the context separates them. A matcher that guessed would flag the friends, and a supervisor sent to read a young person’s private messages for nothing learns to stop going.
What is there instead, today
A profanity filter on every message, images screened before they arrive, patterns tracked over time rather than message by message, and — the part that actually addresses bullying — every conversation readable by the leaders who oversee that young person. The route is a person noticing, not a machine deciding.
Closing this properly needs a language model, a processor agreement, a Canadian privacy review and a cost per message. It is a decision an organization should make deliberately, and we would rather you made it knowing the position than discovered it later.
Who is told
A person, by name, with the thread already open
An alert that lands in a queue nobody owns is a record that something was noticed and nothing was done.
Routed to a named supervisor
Not a shared inbox and not a weekly digest. The alert carries its category, so somebody opening it knows what they are being asked to judge before they read a word of a child’s conversation.
Never to the person it is about
An adult flagged by the scanning cannot see that they were flagged, and neither can they see the trail of patterns building against them. This is enforced in the database, not by leaving a link off a screen.
With the evidence attached
The full thread including anything since deleted, and the audit trail for that conversation, opened together. A supervisor deciding in two minutes decides better than one hunting for context for twenty.
The checking runs on our own servers by matching known patterns of words. No message is sent to an outside company, and no artificial-intelligence service reads your young people’s conversations.
When it cannot answer
The message goes, and the record says it was not checked
Every scanner is sometimes unavailable. What a product does in that minute is a safeguarding decision, and this one is written down.
It sends
If the scan cannot complete — a timeout, an outage, a message it cannot process — the message is delivered rather than held.
Because the other choice is worse
Failing closed means a young person trying to reach a trusted adult is stopped by somebody else’s server being slow. Between an unchecked message and a child who cannot ask for help, this product chooses the first — and does not pretend the first is free.
And it is counted, in the open
The screening log has its own word and its own colour for unchecked, and the header counts how many went out that way. Nobody has to be told the rate; the person responsible can read it.
This is the one place where the product’s general rule — when in doubt, deny — is deliberately reversed, and it is reversed in the direction that keeps a young person able to speak.
In short
What to tell your board
Three sentences, all of them checkable.
- Six behaviours that precede harm are matched on our own servers, and anything found goes to a named supervisor with the whole thread and its audit trail attached.
- Bullying between young people and abuse as a category are not detected, and no screen in this product implies they are.
- When the scan cannot answer, the message is sent and recorded as unchecked, so that a young person is never cut off from an adult by an outage.