Roblox Shares 3 AI Child-Safety Models With ROOST as 123 Million Users Heighten Grooming Risks
Updated
Updated · Fox News · Aug 21
Roblox Shares 3 AI Child-Safety Models With ROOST as 123 Million Users Heighten Grooming Risks
3 articles · Updated · Fox News · Aug 21
Summary
Roblox on Aug. 19 contributed updated versions of three child-safety AI models to the ROOST Model Community, aiming to help other platforms detect grooming signals, personal-information requests and unsafe voice chat.
Version 2.0 of its PII Classifier expands language support to 189 from 17 and lifts its F1 score to 90.52 from 63.41 by analyzing conversations in context, including attempts to move children to other apps.
Sentinel, which flags early child-endangerment patterns for human review, accounted for nearly 70% of cases Roblox detected in the 12 months ended Aug. 7; the company also released a new evaluation dataset for outside testing.
Roblox said its open-source voice safety classifier has been downloaded more than 72,000 times since 2024; Version 3 now covers 30 languages and eight violation categories, with 61% recall at a 1% false-positive rate.
The release comes as Roblox faces scrutiny over predators on a platform averaging 123 million daily active users in Q2, while independent researchers studying more than 2 million messages found harmful chats still slipping through moderation.