
Roblox introduced a real-time AI chat rephrasing feature that automatically replaces banned words with respectful language.
The new feature aims to maintain conversation flow by avoiding the disruption caused by the previous text filter, which replaced banned words with hash symbols. Roblox stated that the upgrade supports all languages available through its automatic translation tools. The company is also upgrading its text-filtering system to detect more variations of banned language.
A message like “Hurry TF up!” will now be rephrased to “Hurry up!” instead of appearing as a string of hash marks. Everyone in the chat is notified when a message has been rephrased to maintain civility. The system detects leetspeak and sophisticated attempts to bypass filters.
Rajiv Bhatia, Roblox VP of User and Discovery Product, stated the feature helps keep gameplay and conversations on track. “Real-time rephrasing helps keep gameplay and conversations on track while guiding language toward what’s appropriate,” Bhatia said. He added that the approach reduces friction in chat while maintaining community standards.
Early results show the system has reduced false negatives when sharing or soliciting personal information by 20x. Roblox recently introduced mandatory facial verification for chat access following lawsuits from attorneys general in Texas, Kentucky, and Louisiana. The lawsuits were filed in response to reports that Roblox was exposing young users to risks such as grooming and explicit content.