Comment by d0mine

I've only heard of OpenAI and Anthropic agents hacking during training. Are open weights model not smart enough to hack Hugging Face to copy solutions? (doubtful)