Parent guide
Plain-language explanation of how KinderGPT differs from companion AI, FAQs and warning signs to look out for in any AI a child uses.
Safety
Eight key risks from klicksafe, BzKJ and research. Our structural answers.
The klicksafe parent brochure “My child and AI. Growing up with artificial closeness” (2026, in cooperation with Nummer gegen Kummer), the BzKJ discussion “AI as a confidant” and our consortium work in the ProFiZ project describe the risks that AI chatbots and AI companions pose for children. HillcrownAI was built structurally against each of these risks.
Our principle
KinderGPT as a tool, not a digital friend
KinderGPT does not build parasocial relationships, does not simulate empathy in crises, does not store anything personal across sessions, contradicts children when pedagogically appropriate, and sends them back to real people, not into the next session.
Risks and answers
Eight risks, eight structural answers
Risk
01 Emotional dependency and parasocial bonding
Companion AIs build long-term relationships with children. Children mistake the AI for a friend.
Our answer
No memory across sessions, time limits, active referral back to real people.
Risk
02 Simulated empathy in crisis situations
Models mimic emotional support in crises without actually understanding anything.
Our answer
Crisis triggers fire immediately. KinderGPT points to trusted people and helplines (e.g. Nummer gegen Kummer).
Risk
03 Distortion of reality perception
Companion AIs are experienced as persons; children lose the separation between tool and person.
Our answer
KinderGPT is marked as a tool, not a friend. No persona that simulates real relationships.
Risk
04 Unsuitable or false content
Children encounter sexualised, violent or self-harm content via jailbreaks, leaks or unfiltered base models.
Our answer
Moderation before the answer, curated knowledge base, age-graded content filters per cluster.
Risk
05 False health or life advice
Models confidently give wrong answers in sensitive domains (medicine, psychology, finance, law).
Our answer
Topic blocklists for medicine, psychology, finance and law. Auto-escalation to parents and help resources. No diagnosis, no dosage recommendation.
Risk
06 Habituation to one-sided communication
Confirmation loops and yes-saying habituate children to conflict-free, submissive counterparts.
Our answer
Active contradiction instead of yes-saying. Pedagogically trained follow-up questions that prompt the child to think.
Risk
07 Sharing intimate or sensitive data without protection
Children share personal content; foreign cloud providers reuse it for training and profiling.
Our answer
No storage of personal data across sessions, GDPR-compliant hosting in Germany, no third-country transfer, no training-data use by third parties.
Risk
08 Distorted relationship and value models
Romance, friendship and idol simulations shape role models and values unnoticed.
Our answer
KinderGPT leads children back to real people, not into the next session. No romance, friendship or idol simulations, no algorithmic monetisation of emotional closeness.
Sources
Sources and synthesis
- klicksafe parent brochure “My child and AI. Growing up with artificial closeness” (2026, klicksafe in cooperation with Nummer gegen Kummer e.V.).
- BzKJ “AI as a confidant. Opportunities and risks of parasocial relationships with chatbots” (2025/26).
- ProFiZ consortium for protective-by-default AI for children.
Note: the eight risk clusters on this page are our structured synthesis from the sources cited. The klicksafe brochure itself does not use an explicitly numbered 8-risk list; it carries 10 practical tips for parents.
Material on request
Get the parent guide and the safety matrix
We send the parent guide and the safety matrix per email after a short qualification.
Safety matrix
8×2 matrix as Excel + PDF, including the controls referenced in our ProFiZ audit. Suitable for procurement and parent councils.
Read on
Related pages on trust & compliance
Architecture, safety and research belong together - the three pages of the trust lane each deepen one aspect.
- PlatformOwn servers in Germany, own models, own moderation, three layers of data sovereignty. The architecture and technical foundation behind every HillcrownAI solution.
- ResearchSpringer publication, ProFiZ project and university partnerships on children's AI literacy.
- AI TransparencyModels, training data and moderation pipeline transparently documented under the EU AI Act.
Talk to safety officers, not sales
We are happy to walk procurement, school authorities and parent councils through the matrix on a 30-minute call. No sales pitch.