Skip to content
Sprout Guard

Menu

Urgent help

AI policy

Message checks, image analysis and AI Check examine different things and can all be wrong. Read what each feature does, where content is processed and which limits apply. A finding in an app is not proof or a guarantee of safety.

  • Photos are analysed on-devicePhotos and findings are not uploaded; analysis can be wrong.

    Image analysis in Safe Camera

    Safe Camera uses AI on the device to analyse photos. Model and analysis files are downloaded from the app’s own address. Images, detections and classifications are not uploaded. This is image analysis, not generative AI or a conversation service.

    A result can be wrong

    Models can miss content or misclassify a harmless image. A marking does not establish who a person is or what happened. Review the photo and covered areas yourself before exporting. The check is neither a safety guarantee nor evidence.

    AI Check

    AI Check processes text, posts, image files and text read from images (OCR) on the device without uploading the content. Origin information may provide evidence or clues. An inconclusive result proves neither AI use nor human authorship. The app cannot give a reliable AI percentage.

  • Word lists and approvals are not AISafe Keyboard uses a word list; adults approve content in Safe Browser.

    Messages, conversations and learning

    Safe Keyboard uses a word list and fixed suggestions you can edit. Parent Coach and Learning have written steps, exercises and missions. They do not use an external chat AI. Suggestions may lack context and do not replace professional advice.

    Approvals in Safe Browser

    Safe Browser has no connected AI. An adult must review and approve text before a child can read it. The open web test stores rules, requests and text fingerprints in private Vercel Blob storage in fra1 (Frankfurt). Full fetched pages are not stored. A parent can delete the workspace’s app data using the current code and confirmation. A deletion marker and timestamp remain without automatic cleanup. Local bookmarks, downloaded copies and provider logs are handled separately. Use fictional test data only.

    Checks in Sprout-chat

    Sprout-chat is available in the open web test. The test for two participants has text only. A word list automatically checks text before a message is stored, delivered or displayed. Messages with detected content are held back. Checks can fail and are not a guarantee. The chat does not read SMS, WhatsApp or Snapchat or check other apps. This is a word list, not an AI service.

  • School does not assess pupils automaticallyTeaching about AI is separate from using an AI service.

    School and Institution

    School does not scan student work for AI, keep an AI access log or assess students automatically. It does not block AI during exams or read private family information. Institution is planned. Teaching about AI is separate from using an AI service in the apps.

  • External AI needs a new assessmentPrivate app data must not be automatically transferred to a new service.

    Before external AI can be added

    A new external AI service requires a separate assessment of purpose, data processing, children’s rights and the applicable lawful basis before it is opened. Private app data on the device must not be transferred automatically. Users must know about the service and its data processing before using the feature.

    Limits for a future service

    A future AI service must not help with grooming, manipulation, threats, harassment, pressure to send images, gambling tips, bypassing rules or other harmful behaviour. Sensitive images must not be sent to the service. Report security issues to security@nordicdts.com using example data.