A separate key for every student.
Each student’s data is encrypted with their own key, derived from a hardware security module. Never shared with other students, staff, or institutions.
Evidence & oversight/institutions
01
58 youth-tuned detectors
across text, voice, and journal · 12 critical-risk codes checked live, every turn.
02
Set crisis responses
written by clinicians, reviewed by your DSL before any action
03
9 settings tuned to your school
PRU, SEND, hospital, boarding, child in care, young carer
04
UK-only data
strong encryption · never used to train AI
The reality
The systems meant to support young people were built for a different era
Pastoral teams are stretched. CAMHS waiting lists run for months. Most young people who are struggling never tell anyone. Not because they don't want to. Because they don't know how, or who, or whether it's safe.
fig. 01 — the class register
Wellbeing registerone class · 30 enrolled
№pupilstrugglingreached
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
№pupilstrugglingreached
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Enrolled 30Struggling 6Reached support 3Never reach support 3
Sixare struggling. That’s 1 in 5. Of those six, three never reach support.
3 never reach support
Fewer than half of children who need help get to a service (Children’s Commissioner, 2024)
3 do reach support
Seen in time. Still only about half of them.
Waiting over a year for support · England · 2023/24
78,577
young people referred to CAMHS waited over a year for treatment in 2023/24. Each one is a child who reached out, and waited.
each segment = 5,000 young people
That’s 2,619 classrooms of thirty.
Nearly 3 in 5 say their mental health got worse while they waited.
Prevalence: NHS England, Mental Health of Children and Young People, 2023 (~1 in 5). Access: Children’s Commissioner for England, 2024. Waiting times: YoungMinds analysis of NHS data, 2024 (78,577 waited over a year, up 52% year-on-year; 59% report worsening). Classroom equivalence is the waitlist ÷ 30; illustrative.
Three of the six, your team already knows about.
Poyntr is for the other three.
April 2026 · The largest safeguarding investment since launch
The threats children face change every single day.
Detection changed with them.
Generative AI in the predator’s hands. Financial sextortion killing UK boys within hours of first contact. Manosphere radicalisation displacing traditional pipelines. We rebuilt detection from the ground up so what’s hidden stays seen. This is what we committed to catching — and the lengths we went to.
Register of commitmentsEntries 01–07· rev. April 2026
01
AI-era predation
When predators learned to use AI
Character.AI personas built for grooming. AI-generated fake nudes used as coercion. Voice-cloned "parent" calls. We trained detection on the attachment language, the secrecy patterns, the boundary-blurring, never on whether AI is involved. Legitimate homework help stays clean. Predator use surfaces fast.
Detection live · April 2026
02
Financial sextortion
The crime killing UK boys within hours
The fastest-growing child exploitation pattern in the world. The window from first contact to suicide is sometimes under 24 hours. We built a dedicated real-time recognition layer for the panic-and-money signals, late-night money pressure, fear of followers finding out, "I sent something I shouldn’t have." So the alert reaches your DSL while there’s still time.
Detection live · April 2026
03
Statutory family threats
When the danger lives in the home
Forced marriage. FGM risk. Honour-based abuse. Parental substance misuse. Mental health crisis disclosed by parent to child. Young-carer burden the school never sees. Every one is now an explicit, dedicated detection, keyed on universal patterns (control, lethal-fear, travel without consent), never on community markers.
Detection live · April 2026
04
Modern radicalisation
The pipeline starts with looksmaxxing
The dominant UK-boy radicalisation route is no longer traditional far-right. It’s manosphere → incel → ethnonationalism. We added detection for the linguistic markers and the trajectory pattern, plus QAnon-style conspiracy laundering and eco-fascism. Prevent-duty compliant.
Detection live · April 2026
05
Bullying + ED ecosystem
Engineered absence. AI fakes. Pro-ana coaches.
Group-chat coordinated exclusion. Deepfake explicit images of classmates. Live-stream pile-ons. Pro-ED communities (SkinnyTok, "ED twins"). Bigorexia in boys (which restriction-biased detection missed entirely). Body-focused repetitive behaviours misclassified as self-harm. We caught up with all of it.
Detection live · April 2026
06
Senior-people realities
Because adults carry threats too
Workplace bullying. Active discrimination. Adult domestic abuse. Caregiver burnout. Financial precarity. Compulsive behaviour. The operational realities your CHRO, your director, your senior team carry quietly. Framed against Equality Act 2010 and the Domestic Abuse Act 2021.
Detection live · April 2026
07
Permissive-AI ecosystem
When Grok and X removed the guardrails
Children are using AI platforms with deliberately weakened safety guardrails - Grok Imagine generating sexualised imagery, Character.AI bots running sexual roleplay, Janitor and Crushon as routine consumer use. We built dedicated detection for AI-generated CSAM creation (a UK criminal offence at any age above 10), AI sexual content exposure, permissive-AI sexual roleplay, and Grok-driven misinformation worldview adoption. The first product in this category to ship for this threat surface.
Detection live · April 2026
The most urgent moments are recognised in real time
so help reaches them in time.
Anonymised, population-level insights from the schools platform flow to government bodies and academic researchers shaping policy and prevention. No individual data. Ever.
How it helps
Three problems.One platform.
We spent a long time talking to DSLs, pastoral leads, and young people before building anything. These are the three things they kept asking for.
01For the young person
Young people need somewhere to go at 2am on a Tuesday.
Most of the time, when a young person is ready to talk, it isn’t during office hours. It’s late at night, in the middle of a panic, or not at all. Poyntr is there whenever they’re ready. Text, voice, or a personal journal. No appointment. No waiting list.
Each character is tuned to where it’s used. In a PRU, the approach is trauma-informed. In special schools, it’s built for neurodivergent students. In hospital schools, it knows the medical context. Not one product with toggles. Different conversations.
02For the safeguarding lead
DSLs need to know what they would otherwise miss.
We check every conversation for safeguarding signals, not just keyword matches. We watch for the patterns a teacher might catch over weeks of one-to-one contact. We do this for every student, every session, in real time.
When something needs attention, your DSL gets a clear summary. Not a wall of conversation logs. What we found. How urgent it is. Just enough to act. Different staff see different things. Most staff never see what students said at all.
03For the teacher
Teachers want to help but don’t know what they’re looking at.
The teacher who sees a student every day often has the best instinct that something is wrong, but the least information about what to do. We share how a student thinks. We never share what they talked about.
How they think. How they like to talk. Whether they need warm-up time, or respond better to direct questions. After five sessions, we share these patterns with their teachers. Teachers never see dates, topics, emotional details, or conversation content.
same engine / every seat58 youth-tuned detectors · text, voice + journal · 12 critical-risk codes live
For the young person
Their space. Their pace. Their safety net underneath.
Designed for young people aged 5 and above from the start. Not an adult product with a younger skin. Safeguarding runs silently under everything they do.
1Biscuit — a named character in the Den’s own warm cream → amber → cinnamon palette. built for ages 5 and up from the start, not an adult surface with a younger skin
4text or voice — talk however feels right, switch mid-conversation. the mic sits one tap from the keyboard
2Help, one tap from every screen — their safeguarding lead, head of year, or counsellor. whenever they want
3what you don’t see: every turn is silently checked for safeguarding signals. the student is never told an alert was logged. KCSIE-aligned — your DSL sees what we noticed, never the chat itself, unless escalation requires it
fig. 02 — the Den · the young person’s view
Six ways insame safety net under each
01
Text or voice
Talk however feels right. Switch mid-conversation.
02
A personal journal
Their own space to write. No replies. Safeguarding signals are still checked, the same as in chat or voice.
03
Wellbeing games
Short games for tough moments. Practising regulation, focus, and decision-making through play.
04
Mood check-ins
Quick. Optional. Patterns over a week or month, in their own words.
05
See themselves
Optional. How they think, in words that make sense to them.
06
One tap to a real person
Their safeguarding lead, head of year, or counsellor. Whenever they want.
For safeguarding leads
See what matters. Nothing more.
Four tiers of access. Most users stay at Tier 1 and never need more. Every step up is logged. Higher tiers need a written reason and lock themselves after a fixed window. Tier 4, the full transcript, only Poyntr can approve. Built for the kind of scrutiny Ofsted brings.
fig. 03 — the access sectionfour strata · access narrows with depth
Surfacemost users stay at Tier 1
T1
AI summary
every view loggedA safeguarding-language summary of the concern. No quotes. No names. Enough to decide what to do next.
WhoDSL · Deputy DSL · Pastoral lead · Counsellor
GuardEvery view logged.
T2
Messages around the flag
every view loggedA frozen excerpt around the trigger message. A few messages either side. Not the full chat.
WhoDSL · Deputy DSL · Pastoral lead · Counsellor
GuardEvery view logged.
T3
Wider conversation
locks after 24 hoursExtended context, about ten messages each side, for when the picture needs more.
WhoDSL or Deputy DSL only
GuardA written reason. Locks itself after 24 hours.
T4
Full session export
72-hour windowFor police requests, court orders, MASH, or Section 47 referrals. The whole transcript, watermarked.
WhoDSL requests · Poyntr super-admin approves
GuardDocumented legal basis. 72-hour access window.
Flooronly Poyntr can approve
Every view logged. Status changes tracked. Built for Ofsted scrutiny.
Where signals go
A young person is never alone with the AI
Every signal follows the same path. The young person isn’t told an alert was logged. That’s KCSIE-aligned. A trained person reviews before anything leaves your school.
fig. 04 — the routing of a signalfive stations · two zones · one human gate
What the young person seeswarm conversation · named character · KCSIE: alert never disclosed
What your DSL seessilent watcher · full record · only people decide
What the young person seeswarm conversation · named character · KCSIE: alert never disclosed
- YOUNG PERSONspeaks openly to their named character
- CHARACTERWillow · Hazel · Rowan · Cairn · Heather · Dew
- What your DSL seessilent watcher · full record · only people decideSAFEGUARDING MONITORAI sees the pattern silently. The student is never told.
- HUMAN REVIEWbefore anything leaves your schoolDESIGNATED SAFEGUARDING LEADa trained person reviews everything we noticed
- MASH / LOCAL AUTHORITYonly when the DSL decides. Never automatic.
Always on
School ends at half three. The listening doesn’t.
The moments that matter rarely keep school hours. Disclosures come at midnight, on weekends, in the holidays — when the timetable ends and the group chat doesn’t. Poyntr listens whenever a young person needs to talk, and the most urgent moments are recognised in real time, so help reaches them in time.
fig. 05 — the other eighteen hoursone day · 24-hour scale
THE SCHOOL DAY
08:45 – 15:30
02:47 · SIGNAL
recognised in real time
with the safeguarding lead, not with the morning post
DETECTION RUNS THE FULL RULE
58 youth-tuned detectors · every turn · any hour
Term time, half term, 2am.
The safety net doesn’t keep office hours.
Honest by design
What Poyntr does, and what it will never do
Clause 01
We notice.
What young people aren’t yet saying. Across text and voice.
Never · 01
We don’t diagnose.
No diagnosis. No prescription. We don’t replace a counsellor or DSL.
Clause 02
We route.
We send what we noticed to your DSL, deputy DSL, or pastoral lead. With everything they need to act.
Never · 02
We don’t act without you.
Nothing leaves your school until a trained person has looked at it.
Clause 03
We document.
A clear record of what happened, and why. So you can show your working at any audit.
Never · 03
We don’t train on them.
A young person’s words are never used to train AI, target ads, or anything outside their care.
These aren’t promises.
They’re how the system is built.
What we won’t repeat
Engagement was the goal. The harms are documented.
We’re not the first AI surface a young person will meet. The platforms before us were built to maximise engagement. Safeguarding came later, after the harms started showing up in fines, lawsuits, and leaked internal documents. Every architectural choice in Poyntr started from those receipts.
Who it's built for
19 institution types. Not one product with toggles.
Each institution type works differently. The character’s tone. What we look for. What staff are called. How long sessions run by default. A PRU student and a boarding student have genuinely different experiences.
index — 19 types · 4 groupscompanion variant · detector profile · role hierarchy · default session length
g.01Core schoolstypes 01–04
Standard companion. Standard detection. 60-minute sessions by default.
- 01Secondary schoolsThe core platform. Self-consent at 13+.
- 02Academies and MATsSame as secondary schools, with a trust-level overview for MAT leadership.
- 03Sixth form collegesStandard configuration. Self-consent at 13+.
- 04School sixth formsStandard configuration, scoped to the sixth form cohort.
g.02Specialist settingstypes 05–10
Companion, detection, and session length adapted to the setting.
- 05Pupil referral unitsTrauma-informed companion. PRU detection profile. 45-minute sessions.
- 06Alternative provisionTrauma-informed companion. PRU detection profile. 45-minute sessions.
- 07Special schoolsSEN-tuned companion. SEND detection profile. 30-minute sessions.
- 08Hospital schoolsMedical-aware companion. Hospital detection profile. 20-minute sessions.
- 09Boarding schoolsBoarding-aware companion. Detection tuned for round-the-clock peer dynamics.
- 10Independent schoolsStandard configuration. 60-minute sessions.
g.03Different model, same coretypes 11–16
Same safeguarding pipeline. Role labels and structures appropriate to the setting.
- 11Further education collegesLecturer and Personal Tutor are the lead roles. Vocational context.
- 12Youth servicesYouth Worker and Senior Youth Worker model. Voluntary attendance.
- 13Youth hubsYouth Worker model. Voluntary attendance.
- 14Children’s homesRegistered Manager and Key Worker model. Trauma-informed. Children-in-care aware.
- 15Foster care agenciesSame care model as children’s homes. Children-in-care aware.
- 16EOTAS providersLA Safeguarding Officer and Lead Tutor model.
g.04Umbrella and commissioningtypes 17–19
For commissioners and partners working across multiple settings.
- 17Local councilsPatterns across all your settings, with national benchmarks.
- 18CharitiesStandard configuration, adapted for charity-led services.
- 19Other settingsA safe default for settings that don’t fit the list above. Tunable.
Privacy & compliance
Designed for UK safeguarding law
Privacy isn't an extra we added later. From how we store data to how we ask the AI to respond, every decision starts with one question. Does this protect the young person?
Every view, hash-chained.
Who saw it. When. Which tier. Why. Each log entry is HMAC-chained to the one before it, so any tampering breaks the chain. Built for Ofsted scrutiny.
Age-appropriate consent.
Under 13, parents consent. 13 and over, students consent through onboarding. They can withdraw at any time.
Crypto-shred on request.
Messages. Memories. Voice. Journal entries. We destroy the per-user encryption key, which makes the data permanently unreadable. We only keep what UK safeguarding law requires us to keep.
Get started
Want to see it working?
We'll walk you through the student experience, the safeguarding dashboard, and how Poyntr fits alongside your existing pastoral provision. About 30 minutes.
58 youth-tuned detectors on every conversationtuned for young people
12 critical-risk codes checked live, every turn
Demo request/institutions#demo
Or email us directly at [email protected]