Sentinel flags messages based on its training. It was trained to distinguish between safe messages and those that were previously reported because they violated Roblox's child-endangerment policy.