Samsung’s Inconsistent Galaxy AI Content Filter Is Causing Some Odd Problems

Rita El Khoury / Android Authority

As with all LLMs, finding the right balance for the Galaxy AI content filter was always going to be tricky. Samsung will be keen not to offend while also not wishing to moderate too aggressively. However, we might at least hope for consistency, especially when it comes to sensitive issues, which is why it seems odd that the AI appears to only censor certain descriptions of skin colors.

In a Reddit post on Monday, user P26601 shared a screenshot showing two examples of Galaxy AI checking sentences for spelling and grammar. The phrase “my skin is white” is checked without issue, while the AI refused to check the phrase “my skin is black,” suggesting it may contain inappropriate content.

We wanted to check the veracity of this report, so a member of the Android Authority team tried the same examples. We got the same results, finding that describing skin as white or brown didn’t trigger the Galaxy AI content filter, while the example using the term ‘black’ did. You can see our results in the screenshots below.

Aside from the most perplexing aspect of this inconsistency, it’s interesting that the AI is so aggressive in moderating how anyone might describe their own skin color. As a further test, we tried using the same descriptions for the skin color of a third party. The results below raise more questions than answers.

This time, Galaxy AI was willing to correct “His skin is brown” but labeled the sentences with descriptions of a third party’s skin being black or white as potentially inappropriate. It at least makes sense that there would be more moderation when describing another person — you’re more likely to offend someone else than yourself — but an inconsistency remains.

This isn’t the first time we’ve seen online criticism of the Galaxy AI content filter. Around this time last year, as users were experimenting with the AI on the new Samsung Galaxy S24 handsets, some took to Reddit to question the levels of moderation. In one case, a Redditor was frustrated that they couldn’t use the word ‘poop’ when exchanging messages with their child’s doctor. Another poster looking for help in romance writing found the AI wasn’t willing to describe a girl as ‘hot.’

Both of these older examples could be explained by Galaxy AI seeking not to endorse profanity or misogynistic text, even if some might see these particular cases as an overly aggressive approach to content filtering. This was also earlier in the life of Galaxy AI, and it isn’t clear if any modifications have been made in response to user feedback.

We have reached out to Samsung for comment on the apparent inconsistencies in the Galaxy AI content filter.

Got a tip? Talk to us! Email our staff at [email protected]. You can stay anonymous or get credit for the info, it’s your choice.