Roblox now rewrites profane chat in real time instead of covering it with hashtags

Roblox has started rolling out real-time chat rephrasing, an AI-powered moderation feature that rewrites profanity into milder language instead of simply replacing blocked words with “####.” The company’s own example is “Hurry TF up!” being changed to “Hurry up!,” with the stated goal of keeping gameplay conversations readable without dropping its civility rules.

That is a meaningful change from how Roblox chat has worked until now. The old filter could break the flow of in-experience conversations by turning large parts of a message into hash marks, while the new system tries to preserve the original meaning in cleaner language. Roblox says everyone in the chat is notified when a message has been rephrased, so the edit is not hidden from other players.

The rollout is limited by age checks and chat scope

The feature is not available across every Roblox conversation. Roblox says rephrasing is for in-experience chat, and its developer documentation says it applies to age-checked users in similar age groups or Trusted Connections. That restriction fits into Roblox’s broader chat overhaul, which began requiring age checks for communication features in January 2026 and uses age bands to limit who can talk to whom.

Roblox also says the rephrasing layer works across all languages currently supported by its automatic chat-translation tools. That suggests the company is treating this as a platform-wide moderation upgrade rather than a small English-only test.

Roblox says this is a tightly constrained AI tool, not open-ended rewriting

Alongside the new feature, Roblox says it is improving its core text filters to better catch violative messages and attempts to bypass the platform’s rules. The company is framing chat rephrasing as a narrower layer that starts with profanity, while repeated or more serious violations still fall under its broader safety and enforcement systems.

One of the more important details is that Roblox told developers the system uses an LLM, but said it is being kept highly prescriptive and limited in scope. In other words, this is not Roblox handing free-form message rewriting to a chatbot. It is closer to a constrained moderation tool that tries to keep tone civil without flattening chat into unreadable blocks of symbols.

The bigger question is whether players will experience this as smarter moderation or as another layer of awkward machine-edited speech. Roblox is clearly betting that a softer nudge will work better than a wall of hashtags, but that will depend on how reliably the system preserves intent without drifting into odd or patronizing rewrites.