> other reasons

Maybe research is sufficiently public and simple to reproduce or the next steps of how to improve things are sufficiently obvious to the smart people working on frontier AI.