There is a recurring issue affecting multiple areas of the Roblox platform in which inappropriate or policy-violating content is not automatically detected or removed when presented using obfuscation techniques involving alternative character sets or encoded formats.
Specifically, users have been able to submit and publish content that includes inappropriate language or phrases by using one or more of the following methods:
Unicode homoglyph substitution: Replacing standard Latin letters with visually identical or similar characters from other scripts, such as Cyrillic or Greek (e.g., using е (Cyrillic) in place of e).
Mixed-script input: Combining characters from different writing systems within a single word to evade pattern-based filtering.
Morse code or symbolic encoding: Representing full words or phrases using dot-dash patterns or other forms of symbolic code that are readable to users but not interpreted correctly by moderation systems.
This type of content has been observed in a variety of public-facing areas, including:
User bios and profile descriptions
In-game chat, both through default text chat and custom UI elements
Asset names and descriptions, such as clothing items or UGC uploads
In all of these cases, the inappropriate content is rendered in a way that is understandable to human readers but is not flagged by automated moderation systems at the time of submission. The majority of these submissions are only removed if they are directly reported by other users and reviewed manually, indicating a lack of proactive detection.
The behavior suggests a systemic gap in moderation coverage for text-based content that is encoded or obfuscated. It appears to apply across both the Roblox website (e.g., profiles, item descriptions) and the Roblox client application (e.g., in-game chat), and may be affecting moderation reliability platform-wide.
No specific usernames, assets, or examples are included in this report in accordance with DevForum policies. The issue is general and reproducible using commonly available character substitution methods.
Expected behavior
Expected Behavior:
Roblox should detect and filter inappropriate text regardless of the character set used, as long as the visual or semantic meaning is the same. This includes accounting for:
Unicode normalization (turning accented or full-width characters into standard ones before filtering).
Cross-script homoglyph detection (identifying when users mix character sets to recreate blocked words).
Multilingual support for common filtered terms.
Suggested Improvements:
Implement Unicode normalization in the filtering system before processing.
Add homoglyph mapping to detect visually similar substitutions across languages.
Support multi-language filtering rules, including basic profanity in common languages.
Create a developer-facing Text Filter Preview Tool, allowing devs to test strings and see how filtering will behave in-game across devices and input types.
A private message is associated with this bug report
