This assumes that GitHub Copilot only gets its data from GitHub and that GitHub Copilot is unique among tools.

Remember the old saying: Anything published on the internet stays on the internet.

What prohibits someone from crawling code from other sites and building a GitHub Copilot equivalent?

Considering how ChatGPT style bots are often trained on public websites that is likely to already be true even.