Roblox Expands Access to AI Safety Tools for Online Platforms

Roblox has made several of its artificial intelligence safety models available to other online platforms by sharing them with the nonprofit ROOST Model Community. This development is intended to support broader adoption of safety tools across the digital industry.

The company’s open-source toolkit includes the Sentinel system, a PII (Personally Identifiable Information) Classifier, and a voice safety classifier. These models are designed to help detect early signs of child endangerment, prevent the sharing of sensitive information, and moderate voice chat for policy violations in real time.

Sentinel, one of the released models, is built to identify patterns in chat that may indicate potential child endangerment before explicit grooming occurs. The PII Classifier uses context from conversations to recognize attempts to share personal details or move discussions off the platform. The voice safety classifier, available since 2024, analyzes speech across multiple languages to flag violations.

Roblox’s safety models, including Voice Safety, PII Classifier, Roblox Guard, and Child Safety models, have been open-sourced for industry use since the end of 2025. The company joined the ROOST Model Community as a founding member in 2025, alongside other technology firms, to promote the development and sharing of online safety solutions.

What the numbers show

• Sentinel identified nearly 70% of detected child-endangerment cases on Roblox in the 12 months ending August 7, 2026.
• The PII filter processes about 370,000 requests per second at peak usage.
• Roblox’s voice safety classifier moderates chat within approximately 15 seconds across eight languages.
• The PII filter achieved a 30% reduction in false positives and a 25% increase in detected PII mentions.

Roblox’s AI moderation systems operate at a large scale, processing billions of chat messages daily. These systems are designed to detect policy violations efficiently, including the sharing of personal information and inappropriate content in both text and voice communications.

Independent researchers have reviewed over two million chat messages on Roblox and identified instances of grooming, sexualization of minors, bullying, violence, and sharing of sensitive information that were not caught by previous moderation systems. The new open-source models aim to address these challenges by providing more advanced detection capabilities.

The open-source safety toolkit is intended to assist other companies, particularly those without the resources to develop their own advanced safety systems. By making these tools available, Roblox seeks to support industry-wide efforts to improve online safety for users, especially children.

Roblox announced the release of three updated AI safety models to the ROOST Model Community on August 19, 2026. These models are now accessible for study and potential adoption by other platforms seeking to enhance their safety measures.

* This article is based on publicly available information at the time of writing.