Roblox already uses AI to moderate voice chats, but the final decision is still made by humans
Anyone who has ever played competitive online games knows how toxic a place voice chat can be in games. Unlike text chat, where it takes time to write something, voice chat happens instantly, making it very difficult to moderate. To solve this problem, Roblox is actively implementing artificial intelligence (AI), but recognises that human control remains necessary to ensure a safe environment for players.
Here's What We Know
During Game Developers Conference (GDC ) 2025, Kiran Bhat, Senior CTO at Roblox, and Hannes Heikinheimo, Head of Voice Chat Security, spoke about the year of using AI for moderation. Bhat noted that the main difficulty lies not only in detecting offensive words, but also in understanding the tone, intensity, and context that determine whether statements are truly offensive.
Despite these challenges, artificial intelligence has proven to be highly effective. Bhat explained that 85% of toxic behaviour boils down to just four main categories, and most violations are associated with about 50 keywords. Machine learning does an excellent job of identifying these patterns, significantly reducing the number of complaints about abuse in voice chat.
However, AI has its limitations. Heikinigheimo emphasised that while algorithms provide stability and impartiality, they cannot always correctly interpret rare or contextually complex cases where human judgement remains indispensable.
Source: Videogamer
