Roblox makes three of its AI safety tools open source

Roblox is releasing three of its AI safety tools to the open-source community through ROOST, the Robust Open Online Safety Tools nonprofit it co-founded with Google, OpenAI, and Discord in 2025. The models include the PII Classifier, Roblox Sentinel, and the company’s latest voice safety classifier, all of which are already running on Roblox’s own platform.

For developers building social features, UGC platforms, or live voice systems, the practical value is obvious: these tools are meant to catch attempts to share or solicit personally identifiable information, spot early signs of potential child endangerment, and moderate voice chat in real time. Roblox is also sharing an evaluation dataset for the PII tool, built from synthetic English-language multiuser chat samples designed around common evasion tactics such as misspellings, coded language, and disguised references to other platforms.

Roblox is also publishing a broader evaluation dataset for safety tooling, with support for moderation in 30 languages and eight violation categories. The company says the goal is to give other platforms a strong starting point for training and tuning their own classifiers, while also benefiting from feedback and improvements from the ROOST community.

The timing matters. Earlier this week, the US Senate Judiciary Subcommittee on Crime and Counterterrorism launched an investigation into Roblox, alleging the platform may be prioritizing revenue and engagement over child safety. Roblox has continued to add safety measures in recent years, including tighter direct messaging rules for...