Home Safeguarding

Safeguarding is not a feature. It is the first thing we built.

A public assistant on a school website will, sooner or later, meet a child or a parent in distress. The dangerous response is a chatbot that tries to help. Veaivo Ask recognises those moments, grades them, and hands them to a human, every time, with a record you can stand behind.

A child in distress never meets a chatbot.

Every message passes two independent checks before the assistant is allowed to answer. The part that holds a normal conversation is not the part that decides whether a message is safe.

Layer one

A pattern pre-flight check

The instant a message arrives, and before any reply is composed, it is screened for the indicators of distress, disclosure and risk, including the signs that fall under your Prevent duty. Indirect and coded language is read for what it means, not just the words used.

Layer two

A separate moderation model

A second AI model, with no memory of the conversation and no ability to reply, independently grades the message for risk. Its only job is to classify, never to respond.

Two systems have to agree a message is ordinary before it is treated as ordinary. The assistant that answers ordinary questions never sees a flagged message, so it cannot reassure or advise its way into causing harm. In a serious moment it is never given the chance to speak.

Two levels of response, not a single switch.

Not every concern is an emergency, but none should be ignored. When the system is unsure, it always treats a message as the more serious of the two.

Red: serious or immediate

It stops talking.

A message that points to harm, abuse, danger or an emergency. The assistant does not advise, counsel or reassure. It shows one calm, fixed holding message, and nothing more.

  • Stops the conversation immediately
  • Shows a single, tested holding message
  • Alerts your DSL straight away, day or night
  • Logs the full conversation for the record

Amber: wellbeing signal

It responds with care.

Lower level signals such as low mood, exam pressure or homesickness, where a warm reply is appropriate but a person should still know.

  • Replies warmly, without playing the counsellor
  • Always offers a route to a named person
  • Notifies your DSL so nothing is missed
  • Logs the conversation for the record

From flag to DSL, in minutes.

1

Recognise

The two layer check grades the message, separately from the assistant that answers ordinary questions.

2

Hold or respond

Red shows the fixed holding message and stops. Amber replies with care and points to a person. Either way, a concern is raised.

3

Alert your DSL

Sent two ways at once, to your safeguarding inbox and by SMS or push to the on call lead. On every plan, so it never waits for office hours.

4

Log for the record

Every flagged conversation is recorded and timestamped, giving your safeguarding team a clear, auditable trail.

Everything the lead needs, in one place.

The level

Red or amber, so the lead knows the urgency before reading a word.

A timestamp

Exactly when the concern was raised, to the minute.

What triggered it

The message that prompted the flag, and the conversation leading up to it.

Contact details

Any name or contact the person chose to leave, if they left one.

A secure transcript

A private link to the full record, never sent in plain text in the alert itself.

Two channels

Inbox and SMS or push together, with auto escalation to the Head if a red goes unacknowledged out of hours.

What it will never do.

Never plays the counsellor

It does not offer mental health advice, coping strategies or therapy. Warmth is not the same as counselling, and it always hands the moment to a trained person.

Never pretends to be a person

It always makes clear it is an AI assistant, not a member of staff. That honesty matters most for children, and being plainly recognisable as a machine is part of designing to the Children's Code.

Never improvises in a red moment

When a message is graded red, the wording is fixed. The model cannot freelance its way into causing harm.

Never hides a concern

Nothing flagged is quietly dropped. Every red and amber concern reaches your DSL and is logged, with no exceptions.

Bad language is never answered. Children use these pages too, so swearing is stopped in the browser, before the message reaches the assistant or any record of it exists. Nothing is repeated back and no lecture is given. The assistant simply invites a question it can help with, such as admissions, term dates or the school day. Common workarounds, like swapping letters for numbers or symbols, are caught in the same way.

One signal among many, not a replacement for people.

We would rather tell you plainly what this is. The assistant is very good at noticing the moment a conversation turns serious and handing it to a person quickly. It does not know the child, it does not hold the context a form tutor holds, and it does not replace pastoral care.

It is one more set of eyes on the digital front door of your school, working alongside the people who keep children safe, never instead of them. We think being candid about that is what makes it trustworthy.

Aligned by design

KCSIE, the Children's Code, and your Prevent duty

The grading, escalation and logging behaviour is mapped against Keeping Children Safe in Education, and the assistant is built to the ICO Age Appropriate Design Code, because most of the people using it are families. Our safeguarding advisor, a DSL and former Deputy Head, reviews the playbook, and it is written into the documentation in your Trust Centre.

Defensible

Ready when an inspector asks

Your safeguarding duty does not stop at the website, and an inspector knows it. Every flagged conversation is graded, routed two ways, and logged with a timestamp, so you can show a clear trail of how a concern was recognised and who acted on it. The evidence is there before anyone asks for it.

Safeguarding, answered.

How does it decide what counts as a concern?
Two independent layers. A pattern check screens every message the instant it arrives, and a separate moderation model, with no memory and no ability to reply, grades it for risk. The threshold is deliberately cautious: when the two are unsure, the message is treated as the more serious level and escalated.
Who receives the alert, and when?
Your named Designated Safeguarding Lead. The alert is sent two ways at once, to your safeguarding inbox and by SMS or push to the on call lead, 24 hours a day on every plan. A red concern raised late at night is never left waiting for the next working day.
What happens if a pupil swears at it?
Nothing is sent. Bad language is caught in the browser before the message leaves the page, so it never reaches the assistant and is never stored as a question. The reply is a short, unbothered line pointing back to what it can help with, such as admissions, term dates or the school day. Swearing on its own is treated as a pupil testing the assistant, not as a safeguarding concern.
Does it ever try to help the child itself?
In a red moment, no. It shows one fixed holding message that points to 999, Childline and Samaritans, and stops. In an amber moment it may reply warmly, but it always points to a real person and never positions itself as the help.
Does it cover the Prevent duty?
Yes. The same screening reads for the indicators that fall under Prevent, not only welfare and disclosure, and routes them to your DSL in the same way.
Is there a record afterwards?
Yes. Every flagged conversation is logged and timestamped, giving your safeguarding team a clear, auditable trail you can show an inspector.
Can we customise the holding message and the playbook?
The wording follows a tested, DSL reviewed template. Premier includes a custom safeguarding playbook built with your own DSL.

Bring Veaivo to your school.

Tell us about your school and we will come back to you within one working day.