Roblox, the online gaming platform used by an estimated 123 million daily active users, unveiled three upgraded artificial‑intelligence safety models on August 19. The tools are part of the Robust Open Online Safety Tools (ROOST) community, which also includes Google, OpenAI and Discord.
What the new models target
The AI classifiers focus on three risk areas: attempts to obtain personally identifiable information (such as phone numbers, email addresses or social‑media usernames), requests to move a conversation to another app, and early patterns that may indicate child endangerment. By analyzing the surrounding context of messages rather than isolated keywords, the system can catch misspellings, split contact details, or coded language that older filters missed.
Improved accuracy and language coverage
Roblox’s PII Classifier version 2.0 now supports 189 languages, up from 17, and its F1 score – a standard measure of classifier accuracy – rose from 63.41 to 90.52. The voice‑chat safety classifier, now in version 3, covers 30 languages and eight violation categories, achieving a 61 % recall rate at a 1 % false‑positive threshold.
Early detection with Sentinel
The Sentinel system scans patterns across multiple conversations to flag potential grooming before it becomes overtly inappropriate. Roblox reports that nearly 70 % of child‑endangerment cases detected in the 12‑month period ending August 7, 2026, were identified through Sentinel’s early‑warning capability.
Open‑source sharing
Roblox is contributing the updated models and an evaluation dataset to ROOST so other platforms—especially those lacking resources to build sophisticated safety systems—can test and adapt the technology. While the tools expand the pool of available safety resources, Roblox cautions parents that no automated system can replace vigilant supervision.
What parents can do
Parents are encouraged to discuss online safety with their children, emphasizing that requests for additional usernames, phone numbers or private messaging apps should be reported to a trusted adult. The company’s tools aim to assist, not replace, parental oversight.
Original reporting: Fox News (HLL/CB) — read the source article.
