I don't think it's much of a departure from original principles. The original impetus for the FSF was the "closing off" of some previously public code by MIT AI

I don't think it's much of a departure from original principles. The original impetus for the FSF was the "closing off" of some previously public code by MIT AI-lab spinoffs, who hired away developers and withdrew from the previously common practice of everyone sharing the code of the systems that were being run at the lab. That led to his views that: 1) if any software is used by users, its code should also be available to and modifiable by those same users; and 2) anyone who tries to keep users from getting access to the code of the systems they use (or uses legal means to keep them from modifying it when they do get access to it), is the users' adversary. If anything the Affero GPL fits the 1980s motivation better than the original GPL does, since Stallman has always had the view that users of multiuser systems, not only the sysops, should also have full source access. A user of a SaaS app is basically the modern-era version of a non-privileged user of a multiuser Unix system. I'm not sure why that use-case wasn't included in the original GPL, come to think of it. My guess is just that it wasn't in practice possible to do what Stallman objected to at a large scale, given technology of the day. If you developed a proprietary app based on GNU software, installed it as sysop, and then didn't want to share the source, your userbase would generally be limited to one site. In theory you could run something more like today's SaaS: a central Unix system where you sold accounts but never distributed the software. In that case Matlab could incorporate GPL'd components without releasing their code by installing Matlab only on in-house servers. They'd sell people shell accounts on mathworks.com, but would never let the software leave the building. In practice this wasn't done at large scale, and instead any widescale usage required distributing the code to sysops at other sites to install local copies (e.g. Matlab licensed its software to departmental and corporate Matlab servers), at which point the GPL would kick in if the software were built on GNU software. The Affero GPL in a sense formalizes that previous status quo, by saying that the GPL distribution requirements don't apply for in-house usage within one site, but do apply once you're making the software available to users outside the organization (even if "making available" is now via telnet or ssh or vnc or http rather than binary distribution). The law around proprietary software has also been moving in this direction, to consider hosted usage to invoke a distribution-like copyright requirement such as "public performance", even if no binary goes over the wire. For example, if I buy an NES game legally, but then I sell accounts where thousands of people can VNC in to my server and play it on an emulator, that is probably not legit, even though I am technically never distributing the ROM to them (it never leaves my servers). This isn't completely sorted out though.