The example you mention about a browser and JavaScript on web pages is a very different situation. A browser can be used to interpret and render any number of web pages and is not in any way tied to a specific web page.
This situation with Bambu is not at all the same. Bambu Studio and their networking plugin are very tightly coupled, to the point that they share in-memory data structures. There is no generic plugin architecture; it's a one-to-one interface where the AGPL side of the code explicitly names, downloads at runtime, and is versioned alongside that networking plugin. Most of Bambu Studio's major features beyond generating sliced gcode do not work without the plugin.
Yes, carving out a chunk that you wish to keep proprietary and then dynamically loading it into AGPL code to try and circumvent the AGPL license's explicit copyleft features is an AGPL violation. See https://www.gnu.org/licenses/gpl-faq.html#GPLPlugins for one fairly clear explanation of this.
Or you can read the Software Freedom Conservancy's take on this exact controversy, now one day later, in which they are unequivocal that this is an AGPL violation: https://sfconservancy.org/news/2026/may/18/bambu-studio-3d-p...
For example the Nvidia driver gets to ignore the Linux kernel license because even though it links with the kernel, when considered as a complete package it's not a derivative work of the kernel. It is its own product that can be plugged into various kernels such as Windows and Linux, and the small adapter layer for each kernel doesn't change that.
Some German court even once interpreted GPLv2 to prohibit tivoization.
By a common sense view it is not a plugin at all, it’s part of the app that they have structured in a weird way to try to avoid the obvious license violation.
Of course, IANAL and just a random passerby with a SUPER rough understanding.
What you're describing is probably a gray area, the closest example I can think of is the Wordpress plugin ecosystem (GPLv2+).
Now, there is some unclear legality around the distribution of plug-ins to a standalone software package. Are you allowed to distribute a GPLv3 plug-in to a proprietary program, like a GPLv3 plugin for Visual Studio? Maybe. Are you allowed to sell a proprietary plug-in for a GPLv3 software, like a paid plug-in to Emacs? Again, maybe.
However, the chances for both go down significantly if the "plug-in" is distributed by the same company as the software itself, and if the plugin is critical to major functionality of that software.
Apple also doing this for their fork of CrossOver's Wine GPL called Game Porting Toolkit, composed by Wine + propietary D3DMetal DirectX library
Bambu took existing open-source, AGPL slicer software for free from the rest of the community and then has continually snubbed that open community by not giving back, or only giving back when they can maintain ultimate control to later decide to be less friendly to their users.
Sorry, no, they can build their own slicer from scratch (hah, good luck) or play by the license. Their networking plugin is tightly integrated with the AGPLv3 Bambu Studio. GNU's stance on plugins is fairly straightforward, and this is not at all a borderline case where the plugin interface is limited: https://www.gnu.org/licenses/gpl-faq.en.html#GPLPlugins
For example of this case is Apple Game Porting Toolkit which is fork of Crossover version of Wine GPL , Apple modified the code add dependency to their own implementation of DirectX's D3DMetal which is proprietary licensed library. So GPTK is composed with mixed GPL Wine app + propietary D3DMetal.
https://www.tomshardware.com/3d-printing/josef-prusa-warns-c...
The situation is interesting and kind of a grey area. The way they structured the network code as a plugin that downloads after first run is obviously to avoid violating the AGPL license, but if you don’t install it, large parts of the slicer are inaccessible. So it’s not at all like a normal plugin.
Edit: a more direct source. https://x.com/josefprusa/status/2054602354851254330
jarczakpawel@ (on Github) reimplemented their network plugin and was targeted with over-inflated legal threats (like Section 1201 claims about circumventing access control, which it did not do) and forced to take down his implementation. What he did is provide an open implementation of their closed source network plugin based on their open-source, AGPLv3 Bambu Studio.
This was not some random "AI slop post". This was a (seemingly AI written, yes) summary from the creator at the center of the current debacle, who seems to be the best non-Bambu expert we have on Bambu's networking plugin.
Josef Prusa's tweet is not at all equivalent information. Prusa's tweet seems to be little more than some conspiratorial thoughts about Chinese law and Bambu being a Chinese company. Does that play in? Maybe, I have no idea one way or the other, but it says nothing about the AGPL violations or the technical details of how tightly integrated the plugin is with Bambu Studio.
Based on his tweets, Josef seems like a guy incapable of coming up with his own original thoughts.
Rather: Josef Průša is not a native English speaker (his mother tongue is Czech).
I observe this for myself: a lot of my original thoughts are hard to translate into English because they are deeply intertwined with how the German language is built.
From my observation in particular native German speakers are prone to this phenomenon (that their original thoughts are quite intertwined with how their native language works). I had this discussion with educated English, Turkish and Russian native speakers, but they honestly told me that for them this relationship between their mother tongue and their original thoughts is not similarly marked.
Are you joking? Guy has been tweeting in English for almost two decades, that’s not the problem.
Here's the frequency of the " - " pattern in tweets by "josefprusa"
2020: 8 / 761 tweets = 1.1%
2021: 5 / 533 tweets = 0.9%
2022: 9 / 597 tweets = 1.5%
2023: 21 / 450 tweets = 4.7%
2024: 57 / 725 tweets = 7.9%
2025: 23 / 390 tweets = 5.9%
2026: 15 / 136 tweets = 11.0%AI-generated text is worth skepticism, but humans can be idiots and spout nonsense too.
Sure, but anyone posting chains of incredibly long LLM generated tweets can safely be ignored as an utter moron.