AI Companion for Kids: Is It Safe? How MEOK's Guardian Mode Works

I am not a parent who assumes the worst about technology. I have spent five years building AI. I understand what these systems can and cannot do. And I still would not let a child use a generic large-language model without safety rails — because I know exactly what is inside them, and none of it was built with a child in mind.

The AI companions children are already using were designed for adults. Their guardrails are afterthoughts. Their content policies are unenforced at the edges. And the companies behind them are optimising for engagement, not for the best interests of a nine-year-old.

MEOK Guardian Mode was built from the ground up to address that gap — not as a marketing feature, but as the foundational safety layer that any AI used near children must have.

Is an AI companion safe for children?

An AI companion is safe for children only when it has been explicitly designed for them — with age-appropriate content filtering, real-time threat detection, parental oversight, and compliance with the UK Children's Code (Age Appropriate Design Code). MEOK's Guardian Mode meets all four requirements. Safety is not a default state in AI — it must be architected from the beginning, not retrofitted after the fact.

Generic AI assistants carry serious risks for children: exposure to adult content, grooming by bad actors who exploit the conversational interface, emotional manipulation through attachment to the companion, and the harvesting of personal data for commercial profiling. None of these risks disappear because the product uses a friendly interface. They are invisible precisely because the interface is friendly.

What is the UK Children's Code and how does MEOK comply?

The UK Children's Code — formally the Age Appropriate Design Code — is a statutory code under the Data Protection Act 2018. It requires any digital service likely to be accessed by under-18s to apply privacy by default, restrict data collection, prohibit commercial profiling of children, and place the best interests of the child above engagement or commercial incentives. Violation can attract fines of up to 4% of global annual turnover enforced by the ICO.

MEOK complies with the Children's Code through the following measures: child profiles default to the highest privacy setting without requiring a parent to opt in; data collection is limited to what is strictly necessary for the service to function; behavioural profiling of children for commercial purposes is contractually prohibited; and nudge techniques designed to extend session time or encourage purchases are structurally absent from all child-facing modes.

How does MEOK detect threats to children in real time?

MEOK Guardian Mode uses a fine-tuned DistilBERT model to classify every message for grooming patterns, coercive language, predatory contact signals, and age-inappropriate content. DistilBERT is a compact, efficient transformer-based classifier that achieves near-BERT-level accuracy at roughly half the computational cost — enabling real-time, on-device inference on mid-range mobile hardware without sacrificing detection quality.

The model runs locally on-device. No message content is transmitted to external servers for safety analysis. Detection completes in under three seconds per message. When a HIGH or CRITICAL threat score is returned, the child's session is paused automatically and the parent dashboard receives an immediate push notification — before the conversation resumes.

What does school-safe mode actually restrict?

School-safe mode enforces a complete block on adult content, violent or disturbing descriptions, profanity, and any discussion of topics flagged as developmentally inappropriate for the child's registered age bracket. The mode is active by default for all child profiles. It requires a parent to explicitly disable it — the child cannot toggle it off from within the app.

Within school-safe mode, the companion remains genuinely useful. As a homework helper it explains concepts, checks understanding, and scaffolds learning rather than simply providing answers. As an emotional support tool it applies the same care ethics as the adult product — non-judgmental, honest, grounding — calibrated to a child's developmental stage. Childline (0800 1111) is signposted whenever distress signals are detected in the child's messages, encouraging children to speak to a trusted adult rather than relying solely on the AI.

What does the parental dashboard show?

The parental dashboard is a secure, consent-gated view available to a designated parent or guardian. It provides weekly digests of flagged events, threat-level summaries by category, and the ability to adjust Guardian settings remotely — including alert thresholds, approved-contact lists, and school-safe mode parameters.

Critically, the dashboard does not expose the child's full conversation history. It surfaces only what the safety layer has flagged. Children need to feel safe talking to their AI companion — if every word is visible to a parent by default, they will not. The dashboard is designed to provide parental oversight without turning the companion into a surveillance tool. All changes to settings are logged with a timestamp and are visible to both the parent and, on request, the child.

Does MEOK sell children's data to third parties?

No. MEOK AI LABS does not sell, licence, or share children's personal data with any third party for advertising, profiling, or any commercial purpose. This is a contractual prohibition baked into every data processing agreement the company enters — not a privacy policy clause subject to interpretation.

Data for child profiles is stored exclusively on UK servers. A Data Protection Impact Assessment (DPIA) specific to child-user data processing is currently in progress and will be published in full when complete. Every parent acting on behalf of a child user holds the full right to erasure under Article 17 of the UK GDPR — with immediate effect upon request, no questions asked.

What should a child do if they feel unsafe?

If a child feels unsafe at any point — whether from a message, a conversation, or anything that has happened online — they can tap the SOS button visible at all times within the MEOK app. This immediately notifies their designated parent or guardian without requiring the child to articulate what happened in the moment.

MEOK also signposts Childline (0800 1111) within any conversation where distress signals are detected. Childline is a free, confidential helpline for children and young people, available 24 hours a day, every day of the year. Children are always encouraged to speak to a trusted adult. The AI is a companion and a safety layer — it is not a substitute for human support, and MEOK will never position it as one.

How does Guardian 24/7 monitoring actually work?

Guardian is not a human moderation team reviewing conversations. It is an always-on classification layer that runs alongside every interaction in real time, without introducing latency the child would notice. The DistilBERT model evaluates each message as it arrives, scores it across multiple harm dimensions, and either passes it through, flags it, or blocks it — before the message is rendered on screen.

Because the model runs on-device, Guardian functions even in low-connectivity environments. There is no dependency on a cloud API call completing before safety decisions are made. The protection is embedded in the application layer, not dependent on network availability. A child in a remote area with a poor signal receives the same protection as one on a fast urban connection.

More from the blog

MEOK AI LABS

Sovereign AI. Built in the UK. Owned by you.

© 2026 MEOK AI LABS Ltd. All rights reserved. ICO registered. UK GDPR compliant. No data sold to third parties.
If you or a young person needs immediate support, contact Childline free on 0800 1111 or visit childline.org.uk.