Hi @tindall@cybre.space, just read your thread at https://qoto.org/web/statuses/106500375406792092 and I totally agree.

#GitHub Copilot's models are derivative works (and derived works) of the sources used to calibrate them and thus can violate #copyright (and patents?!) encapsulated in those sources.

I have a question though.

You wrote:
```
If you trained it on MPL code, you are in violation of that license.
```

What clause are you referring to?

In this days I'm finalizing the #HackingLicense (see http://www.tesio.it/documents/HACK.txt ) and I was thinking about this specific problem.

Assuming it could be expressed properly in legalese (without turning the license in a software-only one), I'd be very happy to explicitly impose that any work automatically built out of a covered work would be covered too. If you know a license that impose such terms, I would be eager to learn more about it (MPL is not one... o maybe I'm missing something?)