It feels unlikely to me that at least first few models at OpenAI/Anthropic were distilled from each other. If it is happening for newer model generations, I doubt either of them would ever publicly admit to this like Musk has done for xAI.

Indeed, they probably would never admit it, but it's common-sense because why not? It's not even illegal, it makes better models, better competitor analysis... of course Anthropic is fully aware of gpt-5.5 capacity, they must benchmark it 24/7.