We find accounts that endanger children.
An independent detection project for Roblox and Discord. Our system scans, our reviewers confirm, and the organisations we work with act on it.
Detection initiatives we work alongside
A Different Approach to Safety Detection
Initiative X
Relies on traditional keyword matching for profile and outfit detection
Understands context and catches nuances that keyword matching misses
Reviewers manually go through flagged users one by one
Scans and flags users automatically, then human reviewers verify the results
Flagging decisions can be influenced by personal biases of those running the system
Detection based on behavioral data, not personal opinions
Focuses on Roblox users only
Flags Roblox users, Roblox groups, Discord users, and Discord servers
Returns limited data like a confidence score and a reason
Detailed findings with the evidence behind each one, and an appeal attached to every decision
Decisions that can't be questioned or independently checked
Published statistics, an open Discord community, and a documented legal position
Generic one-line reasons that often look copy-pasted across users
Generated descriptions tailored to each user's specific violations
Findings stay inside the initiative that produced them
Findings are served through Rayward, where teams can act on them alongside other sources
All flagged users fall under the same category with no distinction
Multiple flag types like Flagged, Confirmed, Mixed, and Past Offender with clear action guidance
600K+ accounts found through manual review
900K+ accounts identified and growing
Lower operating costs but lower detection accuracy
Higher operating costs to achieve higher detection accuracy
Our findings are served in Rayward
Rayward is the client and API that carries our findings, alongside those of other detection projects. Access is granted per account by application, reviewed by a person, and it is free.