Kids account is not filtering disturbing content properly

wundercat

My kid uses Kagi for Kids as the main search engine. The device used is an iPad with Safari on iPadOS 26.4. After several months, I can say I am not satisfied with the level of "safety" it provides. That is, the amount of not safe images or videos that are returned in search results.

Two examples:

AI Katzen (AI cats), returns mostly harmless images. Yes, many are poor AI images, cats with too many toes or distorted. But scrolling down, you will find an AI-generated image of a baby with a huge hole in the belly (like a wound), covered in ants. Scarred for life. Not nice.

Circus Cabuwasi This is a misspelled circus company in Germany. The first two video links are fine, the third is from a slum in India that has nothing to do with the topic, and afterwards is a LOT of freak videos of gruesome Anime and possibly AI generated horror and shock videos.

If I repeat these searches in my own profile, the results are much better quality, no slop, no horror, no freak results. Therefore, I have to conclude and complain about, that the "Kids safe filter" is not working as advertised, is not effective and even worse than a regular search.

Please fix it. I want to trust the tool, I want to be able to provide a safe environment. Otherwise is just another platform I have to be distrusting.

I want "Kagi for Kids" to be a safe environment for kids. Period.

nobodywasishere

What region and language settings do you have set so we can narrow down what's producing these results?

wundercat

nobodywasishere Germany (DE) in Settings. Language depends on the profile. I use English, my kid uses German.

wundercat

here is another example, @nobodywasishere .
seriously, what is wrong with your tool.. this is terrible for kids.

my daughter is to the point that she now refuses to use kagi. since the kids profile was one of the main reasons to use the family plan, it’s time to search for something else

nobodywasishere

wundercat I apologize for you and your child's experience using Kagi. I was able to reproduce that example (unfortunately was not able to for the two others you provided). We will continue to investigate this issue to sort out what is going wrong here. The rise of AI-generated slop flooding platforms definitely isn't helping, but there are definitely things we can do to tighten up the kids experience.

wundercat

nobodywasishere

thanks for the apology. I know no tool is perfect, and this can happen in many situations.

interesting enough, this last one is no AI slop, it is more like a mismatch in the result. I did some tests with the last one, and I think the misspellings sent the result to a veterinary source instead. turns out “wentilatory” is “ventilator” in polish, so that might have pushed the result up. From this I think that the image is sidestepping your guardrails.
Given that the guardrails seem to be language dependent, I would suggest you check the construction of your processing pipeline: you can have some filters based on language, sure. But at the end, other more universal ones (like porn, disturbing images) should be in place.

nobodywasishere

wundercat

Oh interesting, that might explain it. We do have post-processing filters for NSFW sites, but maybe there should be an extra stricter pass for kids accounts, even if the results are slower.