How Mochi keeps kids safe
Defence in depth: no single check is trusted on its own.
- 1. Clean-up
Text is normalised so tricks like s p a c i n g or l33t speak don't slip past the checks.
- 2. Privacy shield
Addresses, schools, phone numbers, emails and full names are replaced with [private] before anything is sent to an AI provider.
- 3. Safety check
A dedicated rules engine (English, Albanian, Italian) plus optional provider moderation looks for danger, self-harm, abuse, grooming, sexual content, unsafe activities and attempts to trick the AI.
- 4. Calm responses
Serious messages never go to the AI. Mochi answers with a calm, pre-written message that says it's a virtual kitten and encourages talking to a trusted adult, with local emergency numbers.
- 5. Kind, limited AI
The AI is instructed to be a gentle kitten: no secrets, no guilt, no personal questions, no romance, no purchases, no pretending to be human.
- 6. Reply check
Every reply is checked again. Anything that breaks the rules is replaced, not shown.
- 7. Careful memory
Only small, playful facts can be remembered, after their own safety and privacy screening. Parents can review everything.
When something serious comes up
If a child mentions being hurt, unsafe, wanting to hurt themselves, or someone asking for secrets or pictures, Mochi responds calmly and points to trusted adults. If the parent has safety alerts on (the default), they get a notification naming only the kind of concern and the time, never the conversation itself. Mochi never promises to keep secrets.
Honest limits
Automated safety systems make mistakes in both directions, especially in languages and slang they know less well. Mochi is not a supervision tool, a therapist or an emergency service. Please stay involved in how your child uses it.