Roblox makes three of its AI safety tools open source

Roblox has open-sourced three of its AI safety models—the PII Classifier, Roblox Sentinel, and its latest voice safety classifier—via the Robust Open Online Safety Tools (ROOST) Model Community, which Roblox co-founded in 2025 with Google, OpenAI, and Discord. The company also released evaluation datasets to help other platforms assess safety tools, with the latest supporting 30 languages and eight violation categories. The move comes as the US Senate Judiciary Subcommittee on Crime and Counterterrorism launched an investigation into Roblox over child safety concerns, requesting records by August 31, 2026.

Roblox makes three of its AI safety tools open source Platform has shared its tools on the Robust Open Online Safety Tools Model Community ROOST , which it co-founded in 2025 with Google, OpenAI, and Discord Roblox is open-sourcing three of its AI safety models https://about.roblox.com/newsroom/2026/08/roblox-open-source-safety-models-roost via the Robust Open Online Safety Tools https://roost.tools/ Model Community ROOST . ROOST is an independent nonprofit that develops and maintains open-source infrastructure for online safety. Roblox co-founded ROOST in 2025 with Google, OpenAI, and Discord. Roblox is contributing its PII Classifier https://huggingface.co/Roblox/roblox-pii-classifier-v2 , Roblox Sentinel https://github.com/Roblox/Sentinel , and its latest voice safety classifier https://huggingface.co/Roblox/voice-safety-classifier-v3 to ROOST. These models are already deployed on Roblox's platform. The PII Classifier identifies attempts to share or request personally identifiable information PII , including misspellings, coded language, and disguised references to other platforms. Roblox is also sharing an evaluation dataset from this tool, which features "synthetic samples of English-language multiuser chats modelled on techniques often used to evade PII filters." Roblox Sentinel detects early signs of potential child endangerment. The latest voice safety classifier moderates voice chat in real time and incorporates feedback from the ROOST community. Roblox is also releasing a new evaluation dataset to help companies assess the effectiveness of their safety tools. The latest version supports moderation in 30 languages and eight violation categories. "Sharing these models gives other platforms a robust starting point to train and tune their own moderation tools," said vice president of trust and safety engineering Naren Koneru. "In turn, we hope to benefit from the learnings and feedback of other companies as they share their work." "Safety is a shared responsibility – no company can solve it alone. Our intent in contributing to ROOST is to help other companies improve their own safety classifiers. We will continue sharing and learning from the ROOST community as we advance automated detection together. Earlier this week, the US Senate Judiciary Subcommittee on Crime and Counterterrorism announced it had launched an investigation into Roblox https://www.gamesindustry.biz/us-senate-to-investigate-roblox-following-claims-it-prioritises-revenue-and-engagement-over-child-safety , claiming the platform "appears to prioritise revenue and engagement metrics" over child safety. Roblox has been told to provide records about child safety, sexual or physical abuse, bullying, and financial exploitation by August 31, 2026. In recent years, Roblox has introduced several new safety measures. These include restrictions on direct messaging for users under 13, global facial recognition checks https://www.gamesindustry.biz/roblox-rolls-out-global-age-checks-for-any-user-wanting-to-use-its-chat-feature , a "sensitive issues" content tag https://www.gamesindustry.biz/roblox-to-introduce-new-sensitive-issues-content-descriptor for experiences, and enhanced moderation policies https://www.gamesindustry.biz/roblox-adds-further-changes-to-safety-systems-and-parental-controls .