Given that the training approach was open sourced, their claim can be independently verified. Huggingface is currently doing that with Open R1, so hopefully we will get a concrete answer to whether these accusations are merited or not.