AI and children's privacy

What AI does in Sixly

• Claude, Anthropic's AI, writes hints and explanations from a student's typed steps and the checker's results.
• Mathpix, a handwriting reader, turns handwritten lines into math Sixly can check.
• For adults, Claude can also read a handwritten line: when Mathpix can't read it, and while Sixly tests Claude as a cheaper second reader. During that test Mathpix's reading is the one Sixly checks and shows; Claude's is only compared with it.
• AI never decides whether an answer or a step is right. Answer keys and a math checker that Sixly runs (SymPy) decide that, and a hint that contradicts the checker or gives away the answer is not shown.
• Each time a student opens the practice desk, and again after 3 hours, Sixly says it's an AI tutor: software, not a person. With voice on, it also says so out loud.
• With voice on, a student holds a button to ask out loud. Microsoft's Azure speech service turns the question into words, and the tutor answers out loud in a voice from ElevenLabs (or Azure), with captions of every word. The tutor never speaks first.

What AI companies receive

• Every request to Claude goes straight to Anthropic's API. Under Anthropic's terms for businesses, it doesn't train its models on Sixly's requests, and it deletes them within 30 days, unless a safety review or the law needs them longer.
• Requests carry no names or email addresses; a scrambled code stands in for the student.
• Mathpix gets only the pen strokes of one line at a time, with no name, and doesn't save the writing.
• When Claude reads a line for an adult, Anthropic gets a picture of that one line, with no name, under the same terms. Students under 18 never have a picture of their handwriting sent to Claude.
• Azure gets the recording of a question asked out loud, with no name, and keeps neither the recording nor the words. ElevenLabs gets only the tutor's words, never a student's voice.
• Nothing Claude or Mathpix writes is ever used to train Sixly's own models.

Students under 18

• A parent or guardian holds every account for a student under 18, and decides whether AI hints and handwriting reading are on.
• Students under 18 can't type free questions to the tutor. If any typed question reaches Sixly from them, the server drops it before anything else reads it.
• Without AI hints, students still get typed-step checking, grading and Sixly's own written hints.
• AI hints for students under 18 stay off until this statement is published and Sixly has confirmed its safety checks; we'll say here when they open.
• A student 13 to 17 may talk with the tutor only with a parent's yes. Their spoken words show as a caption on their own screen and aren't sent to the AI; the tutor answers with a hint on their work.
• Anyone who types or says a question that suggests they may hurt themselves sees a crisis line card (988), and the question isn't sent to AI.

Children under 13 (COPPA)

Sixly follows the Children's Online Privacy Protection Act (COPPA) for children under 13:

• A parent gives verified consent, by a consent form signed by hand and sent back as a photo or scan, before Sixly collects anything from the child.
• AI hints need the parent's separate yes, and a no changes nothing else.
• No child's handwriting is sent to Mathpix, and children give no email address or phone number.
• Parents can review everything collected from their child, delete it, and stop further collection.
• Sixly keeps a written retention policy and deletes children's data on its schedule.

Accounts for children under 13 aren't open yet. Details are in the children's privacy notice.

For review, needs a lawyer: Anthropic's guidelines for organizations serving minors (updated Mar 16, 2026) ask for age verification, moderation, monitoring, this public statement and AI disclosure (vendor review 2 and 4). Plan 6.4 lists what must be true before SIXLY_MINOR_AI goes on. The keyword crisis screen for adults is open question 3.

Contact

Questions or concerns about AI in Sixly: privacy@sixly.ai.