That is my whole point: MIT and BSD code is free-er than GPL and AGPL because you can go from MIT and BSD to AGPL but not the other way around.
The original code will still be under the permissive license. But you can take it, and fork it, and continue to develop the fork, and all your new additions can be licensed under the more restrictive one (including GPL).
To be clear about what's going on, you could put a notice saying that anything taken from Source Project X was MIT, all further changes were (GPL). Perfectly legal. But really, how would people know which parts are which except diff'ing it, they might as well just go back to the original -- or they can always diff if if they want regardless of license of course. MIT/BSD don't require attribution or anything, they just say do whatever you want with the code.
So forking and just saying "the license is [eg] GPL" is pretty much the same thing. It's legal. You can't stop people from continuing to use the original code without complying with the GPL, but your fork is GPL.
IANAL. Maybe if it ever got litigated it'd end up different. It might not get litigated, for a while.
With BSD code, the majority of devs can always sell out, and make sure all future modifications are purely proprietary.
We’ve seen that happen on Android.
Nowadays neither the launcher, nor the keyboard, nor the dialer, nor the contacts, nor the calendar, nor the device search is open source anymore.
A lot of code that ended up under free licenses has been proprietary first and while they cannot revoke the free licenses from already released code they can make their next release non-free as long as they haven't accepted outside contributions without a suitable contributor agreement.
Not all contributor agreements are equal, and I would recommend choosing which projects you will not contribute to based on their CLAs. If they ask you to give them copyright without good reason, they are preparing to fuck you over. Ironically, this is why I will not contribute to GNU because they require copyright assignment so they have a higher chance of winning conservancy cases (IANAL, but I have one in my family and this is bullshit in my country but may be true in the US, so I feel conflicted about it).
> A lot of code that ended up under free licenses has been proprietary first and while they cannot revoke the free licenses from already released code they can make their next release non-free as long as they haven't accepted outside contributions without a suitable contributor agreement.
The only example I can think of is OpenSolaris (which required contributors to sign a CLA that assigned copyright to Sun/Oracle). Most modern projects don't work like that, and if you have a copyleft license, a company could not legally create a proprietary fork and distribute it.
Besides, if that happens you end up with free software forks (illumos in the OpenSolaris case) and everyone moves to using that.
I guess, I hadn't much considered whether that was a thing one could do.