Roblox has released three of its artificial intelligence safety tools as open-source resources

Two mobile phone screens displaying a virtual environment with avatars and notification pop-ups related to Roblox voice chat policy violations.

Image credit: Roblox

Roblox is making three of its AI safety models available to the public by contributing them to the Robust Open Online Safety Tools (ROOST) Model Community.

The independent nonprofit ROOST focuses on creating and managing open-source online safety infrastructure. Roblox joined hands with Google, OpenAI, and Discord to establish ROOST in 2025.

To aid the community, Roblox is offering its PII Classifier, Roblox Sentinel, and its newest voice safety classifier. These models are already operational on Roblox’s platform.

The PII Classifier helps detect the exchange or request of personally identifiable information (PII) using various methods such as misspellings or disguised references. Roblox is also providing an evaluation dataset from this tool, showing "synthetic samples of English-language multiuser chats" that often bypass PII filters.

Roblox Sentinel identifies early indicators of possible child endangerment. The new voice safety classifier manages real-time voice chat moderation and integrates feedback from ROOST’s community.

A new evaluation dataset is also being released by Roblox to support businesses in evaluating their safety tools. The upgraded version moderates 30 languages and covers eight violation categories.

According to Naren Koneru, Vice President of Trust and Safety Engineering, "By sharing these models, other platforms have a strong basis to develop and refine their moderation tools." He added, "We aim to gain insights from the industry and learn from their contributions as well."

Furthermore, Koneru stated, "Safety is a collective duty; no single company can tackle it alone. Our goal in contributing to ROOST is to enable other companies to enhance their safety classifiers. We will persist in sharing insights and learning within the ROOST community as we progress in automated detection."

Recently, the US Senate Judiciary Subcommittee on Crime and Counterterrorism disclosed an inquiry into Roblox, alleging the platform emphasizes revenue and engagement above child safety. Roblox has been instructed to submit records concerning various issues relating to child safety by August 31, 2026.

Over the years, Roblox has implemented numerous measures to boost safety. These include restricting direct messaging for those under 13, rolling out global age verification, introducing a “sensitive issues” content label, and enhancing moderation policies.