Transparency report
We do not ask you to trust adjectives. These are the live numbers from Kyloen's child-safety system for the last 90 days, and the honest status of every safety layer.
Updated 12 October 2026. Aggregate counts only; numbers below 5 are shown as "fewer than 5" so no child can be identified.
Last 90 days
- Messages flagged by the classifier
- 0
- Flags reviewed by the second-stage AI
- 0
- Flags cleared as false alarms by that review
- 0
- Parent alerts raised
- fewer than 5
- …of which urgent
- 0
- Human reviews completed
- 0
- Human reviews that found a false alarm
- 0
The 12 safety layers
1. Policy as code
LiveKylo's safety rules live in one versioned prompt, tested before every release.
2. Input screening
LiveEvery child message is checked by an English and Hinglish classifier.
3. Constrained generation
LiveFixed, cached safety instructions on every reply; age-specific limits.
4. Output screening
LiveReplies are checked before a child sees them (respectful tum-forms, no tech talk, length limits that never cut helplines).
5. Three-stage crisis review
LiveClassifier, then a second AI review with context, then a human for every urgent case.
6. Silent alerts
LiveThe child is never told an alert was sent; parents see the concern, never the child's words.
7. Human on call
LiveEvery urgent, harm-at-home or disputed alert goes to the safety team with a 60-minute review target, reminders every 30 minutes, and a recorded human decision.
8. Data minimisation
LiveRaw chats are deleted after 90 days.
9. Consent and parental control
In progressA parent declares they are 18 or over, verifies their email with a one-time code, and every consent is recorded with date and policy version. Government DigiLocker-based verification needs Kyloen's onboarding as a partner and is planned.
10. No ads, no tracking of children
LiveNo ads, no selling data, no analytics inside the child experience, no training on children's chats.
11. Evaluation, red team, audit
In progress49 scenarios and 27 crisis cases gate every release, and an internal red team (an attacker AI and a judge AI) probes Kylo; its finds become permanent tests. An independent external audit is the remaining step, with an audit pack ready.
12. Transparency reports
LiveThis page, updated daily.
Questions or a safety concern? Write to safety@kyloen.com. If a child is in danger now, call 112 or Childline 1098.