If this is the case, then the system needs to be far less complex. The algorithm should absolutely be tuned in the direction of steering children toward content suitable for children and away from that which is not. They know this is not currently the case and doing nothing, so yes, they're purposely keeping it harmful.