They are probably doing it to limit AI LLM bots from hoovering up the code they’ve already hoovered up.
Why is this a bad thing, when M$ is already training on it themselves? If your code is permissively licensed, it seems logical or even desired to be scraped for LLMs