Unfortunately the article is wrong, the OpenAI deal relents on exactly what Anthropic wouldn't. The OpenAI deal is "you can do anything that's 'legal'". This is the same as "zero restrictions", in the real world. Anthropic was offered the same and said "no, these two usecases are not allowed, period".
I had been finding it super unsettling: Like, why would DoD go nuclear over wanting to build fully autonomous weapons… with Claude?
But no, it’s just run-of-the-mill Trump administration corruption. Phew!