Responsible AI at KidsAI

Last updated: August 2026

This trust document is maintained in English. See also: Security & Incident Response · Privacy Policy · About & Safety.

1. What KidsAI's AI does — and deliberately doesn't do

KidsAI turns a child's own drawing into styled artwork and short animations. The AI's job is to celebrate the child's idea, not replace it. That single principle drives most of our design decisions:

The child's artwork is always the starting point. Every generation begins from a drawing the family uploads. There is no blank-canvas "type anything" image generator.
Curated styles instead of open prompts. Families choose from a reviewed set of art styles. Children never type free-form prompts into the image pipeline, which removes the main pathway to unsafe or age-inappropriate output.
No open-ended chat. KidsAI has no chatbot and no conversational AI a child could talk to.
Adult-guided by design. Accounts are created by adults; our teacher materials and product flow keep upload, style choice, and download in adult hands.
AI output is framed as a transformation, not an "improvement." Our classroom resources explicitly position the AI result as a comparison and storytelling object next to the child's original — never a "better version" of it.
No ads, no selling data, no training on children's uploads. Uploaded drawings and generated works are not used to train KidsAI's models or our AI providers' models (we use paid APIs whose terms exclude training on submitted content).

2. AI-generated content is labeled

Wherever KidsAI shows an AI result — the coloring result screen, the animation screen, and saved works — the work carries an "AI-styled from your drawing" label. The label exists so that a generated image is never mistaken for the child's own hand-made work, which matters for classroom use, family sharing, and honest media literacy.

3. People stay in control (override & agency)

Users can override or reject every AI decision in the product:

Choose or change the style — the AI never picks a final style for you; for animations, the "AI auto-match" mood is an optional recommendation that users can override by picking any mood manually.
Regenerate — try the same drawing again, or with a different style, if the result misses.
Delete — remove any generated work (and the upload behind it) from your account at any time.
Report — flag any result directly in the product (see below).

4. How we monitor for bias and safety issues

Generative models carry the biases of their training data. We assume issues will appear, and we build the process around catching them:

Pre-launch style review. Every new art style is tested against a matrix of real children's drawings (varied ages, subjects, and drawing skill levels) — never a single sample — before it ships.
Production failure logging. Failed or degraded generations are recorded with error categories, so silent quality regressions surface in data, not just complaints.
Founder review of outputs. Generated works are routinely reviewed for quality, fidelity, and safety issues, with per-work review notes used to track and follow up on problems.
Input safety moderation. Images submitted for animation pass a child-safety content check before any AI processing; rejected inputs are never charged.
User reports feed the loop. Every in-product report is reviewed, and confirmed issues become prompt-level or model-level fixes with regression tests, so the same bias doesn't quietly return.

5. Real examples: bias we found and fixed

Transparency means showing our work. Two real, recent examples from our monitoring process (sanitized of internal details):

CASE 1 · AESTHETIC NORMALIZATION

The model kept "redesigning" children's characters toward generic, polished looks

Found (July 2026): in some painterly art styles, our review caught the model replacing a child's original character — its proportions, invented features, and quirky choices — with a more "professional-looking" generic character. That is an aesthetic bias: the model was treating mainstream character design as the standard the child's art should converge to, erasing exactly what makes a child's drawing theirs.

Fixed: we added a mandatory fidelity requirement to every style's generation template — the child's original character design, proportions, and creative choices must be preserved; the style may change the medium, never the idea. New styles cannot ship without this clause, and the fix was verified against our regression set of real children's drawings and extended with an audit across existing styles.

CASE 2 · LANGUAGE BIAS

Animated characters spoke English by default, whatever language the child speaks

Found (July 2026): our animations can include short spoken lines. The generation pipeline defaulted that dialogue to English — an anglocentric default that told non-English-speaking children their language wasn't the "normal" one for their own characters.

Fixed: dialogue language now follows the child's world first: if the drawing itself contains written words, characters speak that language; otherwise they speak the family's interface language (KidsAI supports eight). English is a fallback, not the default.

6. Report a problem — directly in the product

Every result screen and saved work in KidsAI includes a "Report a problem" option with these categories:

Unsafe or inappropriate output
Biased or stereotyped result
Privacy concern
Something else

Reports go straight to the operator with the affected work attached, are reviewed against the monitoring process above, and materially change the product — that is how several of our fixes started. You can also email aistudio4kids@gmail.com anytime.

7. Known limitations, stated honestly

Generative AI can still make mistakes: odd anatomy, missed details from the original drawing, or a style that lands differently than expected. The regenerate / change-style / delete controls exist precisely for this.
Bias monitoring is a process, not a finish line. We publish fixed cases here as they are confirmed, and we expect this page to keep growing.
AI output is a creative transformation. It should sit next to a child's original work — our classroom reflection activity shows how we recommend framing that conversation.

8. Questions and feedback

We welcome scrutiny from parents, teachers, and reviewers: aistudio4kids@gmail.com. Related reading: Security & Incident Response · Privacy Policy.