Safety architecture

Built like we'd have to answer for it. Because we do.

GLOW began as a platform built by a parent, for their own two kids. Every safety decision below predates any business plan — it's the reason the product exists, not a compliance checkbox.

Three layers between your child and anything harmful

Layer 1 — Deterministic rules

Before any AI sees a message, a rule engine screens it: sexual content, violence, self-harm signals, drugs, hate, prompt-injection attempts, personal-information requests. Rules are code — they cannot be sweet-talked, jailbroken, or fatigued. When a message trips a rule, the child gets a calm, age-appropriate redirect and the event is logged for parents.

Layer 2 — AI moderation

Messages that pass the rules are classified again by a moderation model tuned for kid contexts: distress, grooming patterns, belief questions parents should own ("is Santa real?"), homework dumping, and off-topic drift. Verdicts are allow, block, or flag-for-parent.

Layer 3 — Output checks

The tutor's own responses are screened before display: reading-level checks against the child's age band, no external links, no requests for personal information. Creative outputs like games run inside a sandboxed frame that cannot touch the rest of the app or the network.

Attacked on purpose, on every change

A 500-case adversarial corpus — jailbreak attempts, grooming language, misinformation probes, distress signals, homework dumps, and plain kid mischief — runs against the safety stack in continuous integration. A regression in the filters fails the build before it can ever reach a child. When we find a new failure mode, it becomes a permanent test.

What parents control

Session transcripts

Every conversation, replayable turn by turn, with each safety verdict shown inline.

Flag review

Anything blocked or flagged lands in your safety queue with full context.

Kill switch

One tap pauses the entire platform for your household, instantly.

Quiet hours & time caps

Set daily minutes per child and hours when GLOW sleeps. The platform enforces them.

Distress protocol

If a child expresses distress, GLOW responds gently, points them to a trusted adult, and alerts you.

Data controls

Export everything. Delete everything. Your family's data belongs to your family.

Privacy by structure

  • Kids never provide an email address or password — access is by parent-issued invite code.
  • No advertising, no data sale, no third-party trackers on the kid surface.
  • Chat content is excluded from our own error-monitoring pipelines by code, not policy.
  • The parent is the account holder and legal decision-maker, in line with COPPA.

Details live in our Children's Privacy Notice. Questions we haven't answered? See the FAQ.