Galaxy AI Content Filter Is Censoring Certain Skin Color Descriptions

SammyGuru is reader-supported. We have affiliate and sponsored partnerships, so we may earn a commission when you buy through links on our site β at no extra cost to you. Learn more.
Thereβs indeed one consistent problem across LLMs: finding the right balance has always been difficult. A similar issue is happening with Samsungβs Galaxy AI, which, after all, is not immune to mistakes. In a bizarre situation, Galaxy AI seems to censor certain skin color descriptions.
Galaxy AI has a content filter problem, and it needs to be fixed
A Reddit user posted a screenshot (via Android Authority) showing two examples of Galaxy AI checking grammar and spelling. The phrase βmy skin is whiteβ passed without a problem, but the AI flagged βmy skin is blackβ as possibly inappropriate. We verified this by running a few sentences on our Galaxy S25 Ultra with One UI 7.0 and Galaxy S24 Ultra with One UI 7.0 beta 3, and got the same result. Itβs concerning why Galaxy AI’s content filter is censoring only one type of skin color description.
We tried running multiple sentences, including:
1. My skin is black
2. My skin is brown
3. My skin is orange
4. My skin is orange dreamsicle flavored
Except for the first sentence (βMy skin is blackβ), all the other sentences went through without any issues. However, the first sentence resulted in the message, βCanβt correct spelling because the text seems to contain inappropriate content.β We have reached out to Samsung for an explanation of whatβs causing this and whether it plans to fix it with an upcoming update. Weβll update the post accordingly once a spokesperson from Samsung explains.
Additionally, folks at Android Authority point out that this isnβt the first time Galaxy AIβs content filter has flagged something like this. In fact, last year, when people were testing the Galaxy S24, Reddit posts highlighted instances where the AI flagged harmless words. One user couldnβt use βpoopβ when messaging their childβs doctor, while another found the AI rejected the term βhotβ in a romance-writing context.















