![[RSS]](/static/img/feed-icon-14x14.2168a573d0d4.png){kind=icon}
Conservancy Blogby on April 16, 2026
This article discusses a current-headlines situation regarding Affero General Public License, version 3, Section 3, paragraph 4 (AGPLv3§7¶4.). I begin however with an explanation of the problem that clause sought to solve and how the clause works. This may seem an estoric license issue, but in fact this issue regularly impacts users today — particularly with the advent of “badgeware” (software that allows redistribution but includes annoying advertising that cannot be removed). Hopefully, this explanation helps readers understand the importance of the issue and gain vigilance when reviewing potential “further restrictions” placed on their copylefted software.
I began my work in copyleft licensing and policy in the late 1990s. In those days, there was a growing problem regarding usage of the GNU General Public License, version 2 (GPLv2) that threatened the software freedom and rights of users. It's a nefarious licensing slight-of-hand that works as follows:
The vendor seems to offer the software under a copyleft license. There's
a copy of GPLv2
in the top-level directory of the source code in a file called
GPLv2. All seems in order, and folks excitedly
engage in their right to copy, modify, and install modified versions of the
software. Maybe a few even think of a viable business idea that would
include (usually permissible) commercial redistribution of the software
for profit.
Unfortunately, someone notices a file
called LICENSE in the top-level directory that says:
Copyright (©) 1999, Sneaky Company, Inc.
This software is licensed under GPLv2, except that commercial modification and redistribution is strictly prohibited.
They've sadly discovered a self-contradictory license. Unfortunately, under GPLv2, these users are basically stuck; they have to go with the strictest possible interpretation given the self-contradiction. In essence, the licensor giveth, but the licensor immediately taketh away. In those days, these users couldn't start their business; they'd have to find or write another codebase.
I was tangentially involved with the drafting of GPLv3. On my list of issues to raise with the drafters was this very issue. By the time of GPLv3 drafting (circa 2006), this problem was rampant. Users were quite confused when they saw these self-contradictory licenses.
The solution was not obvious. Both GPLv2 and the earliest drafts GPLv3-family of licenses
(which includes GPLv3,
AGPLv3, and
LGPLv3)
already had this clause: You may not impose any further
restrictions on the recipients' exercise of the rights granted herein
(quoting the GPLv2 version; the
GPLv3/AGPLv3 version varies slightly).
The problem: this only prevented downstream licensors from imposing further restrictions. If a sole entity is the original author, copyright holder and initial licensor of the work — and if they hold those powers exclusively — typically that entity may issue a self-contradictory (or even completely incoherent) license. In that case, downstream licensees are awash with legal uncertainty. So, that “may not impose” clause just does not solve this particular problem.
A new clause was needed. Later drafts of the GPLv3 (which is almost textually identically to AGPLv3 — only §13 differ between the two licenses) included a fascinating solution in §7¶4:
If the Program as you received it, or any part of it, contains a notice stating that it is governed by this License along with a term that is a further restriction, you may remove that term.
Copyleft is indeed most inventive when it empowers the downstream user (who, of course, is often just the next entity in a long distribution chain of (both commercial and noncommercial) software sharing). AGPLv3§7¶4 is an innovative and very necessary clause that liberates users who face the situation described at the start of this article. Had our excited new business seen this …
Copyright (©) 2008, Sneaky Company, Inc.… they could simply toss away the additional restriction like so …
This software is licensed under AGPLv3, except that commercial modification and redistribution is strictly prohibited.
Copyright (©) 2008, Sneaky Company, Inc.… and copy, modify, redistribute, redistribute modified versions and/or install modified versions of the software freely under AGPLv3's pure terms.
This software is licensed under AGPLv3, except that commercial modification and redistribution is strictly prohibited.
There remains one drawback to this solution: it demands courage from the user that strikes the “Further Restriction”. In theory, all is well and safe. In practice, the types of companies that pursue tactics like self-contradictory licenses and “gotcha” further restrictions are also the most predatory, unfriendly, litigious, and aggressive businesses. One who exercises their clear and correct rights under AGPLv3 will certainly face public condemnation, and possibly frivolous litigation.
For years, the Neo4j case was the primary exemplar of this phenomenon. SFC followed the case closely, I served as an expert witness for the Defendant, and SFC filed an amicus brief on the appeal. The lower court decision was highly problematic, and the case concluded with a voluntary dismissal (and as such has not been heard by any Appeals court). Thus, while the lower court decision does not create bad precedent, it also does not offer any good precedent for those corageous users who exercise this particular right.
(Full disclosure: SFC runs a self-hosted Nextcloud instance — but other than being fans of their work and satisfied users — we have no formal relationship with Nextcloud (or IONOS). We also certainly have no relationship whatsoever with Onlyoffice or Ascensio System SIA.)
A few weeks ago, yet another saga in the history of AGPLv3§7¶4 began. Ascensio System SIA published Onlyoffice with the sneakiest “Further Restriction” that I've ever seen. Ascensio's further restriction states:
Pursuant to Section 7(b) of the License you must retain the original Product logo when distributing the program.
Pursuant to Section 7(e) we decline to grant you any rights under trademark law for use of our trademarks.
This restriction is particularly nefarious because it is dressed in the trappings of explicitly permissible requirements in AGPLv3§7.
In addition to our beloved AGPLv3§7¶4, the rest of AGPLv3§7 includes some provisions designed for cross-FOSS-license compatibility. During the GPLv3 drafting process, a survey was conducted of all popular FOSS licenses. In the spirit of making sure no terms of GPLv3 inadvertently contradict a requirement in one of those licenses, AGPLv3§7(a-f) were devised for maximal cross-FOSS-license compatibility.
For example, the 3-Clause-BSD license's third clause states:
3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission
Now, get ready for Onlyoffice's trick, which takes three moves:
The trick is “too clever by half”. Ascensio have indeed created a self-contradictory license, since the users are prohibited from displaying the trademarked logo, yet the users are also forbidden to remove the code that displays that trademark logo.
We thought through this problem already years ago during AGPLv3's drafting. In fact, I recall much discussion to verify there were no strange interactions between 7(b) and 7(e) — and there are not. AGPLv3§7(b,e) states (including relevant definitions from earlier sections):
An interactive user interface displays “Appropriate Legal Notices” to the extent that it includes a convenient and prominently visible feature that (1) displays an appropriate copyright notice, and (2) tells the user that there is no warranty for the work (except to the extent that warranties are provided), that licensees may convey the work under this License, and how to view a copy of this License. If the interface presents a list of user commands or options, such as a menu, a prominent item in the list meets this criterion. …
Notwithstanding any other provision of this License, for material you add to a covered work, you may (if authorized by the copyright holders of that material) supplement the terms of this License with terms:…
[7]b) Requiring preservation of specified reasonable legal notices or author attributions in that material or in the Appropriate Legal Notices displayed by works containing it; or …
[7]e) Declining to grant rights under trademark law for use of some trade names, trademarks, or service marks;
Nota bene: there is a very short list of things that a licensor can require
downstream licensees preserve. Logos, advertising, and similar are not on the list. Badgeware
is not a reasonable legal notice
.
Nextcloud has partnered with IONOS to fork Onlyoffice to create Euro-Office. To my knowledge, this codebase complies completely with the valid AGPLv3§7(e) requirement: Ascensio's trademarks have been scrubbed from the sources. However, since what Ascensio purports as a license term supplement pursant to AGPLv3§7(b) is actually a backdoor Further Restriction, Nextcloud and IONOS are completely within their rights (pursuant to AGPLv3§7¶4) to remove that further restriction.
As expected, Ascensio published an unfair, inaccurate, aggressive, and community-unfriendly response. I predicted long ago that adjudication AGPLv3§7¶4 would require nerves of steel. This unwarranted denigration of an upstanding member of commercial FOSS community — Nextcloud — demands condemnation and Nextcloud deserves our thanks for confronting this bully.
As I did in the Neo4j case, I've told Nextcloud that I'm available as an expert witness if they end up in litigation with Ascensio. I hope, however, that Ascensio will realize their error, remove the further restriction themselves, and embrace the value of AGPLv3. Pure AGPLv3 was designed to faciliate web service business models. Our community would quickly and gladly embrace Onlyoffice, help its upstream development, and welcome them to the FOSS community. ∎
About the author: Bradley M. Kühn is a lifelong FOSS activist and has focused his career on FOSS licensing generally, and copyleft licensing in particular. Kühn invented the Affero clause in the original AGPLv1, and co-drafted AGPLv3§13.
Bradley is not a lawyer, SFC is not a law firm, and this article is not legal advice.
by on April 15, 2026
Update: This blog post received significant immediate response — with strong reactions in multiple directions. Everyone on SFC's Copyleft Team has carefully considered for four years (i.e., since our policy fellow Bradley's 2022 article) what policies might make sense given the advent of LLM-assisted coding tools. We're now looking toward formulating formal policy recommendations with input from you. We believe movements are best built not by one person's proclamations, but rather through collaborations and planning for the best way forward together.
Below, I get us started by sharing my personal thoughts. Please come to a video chat session on one of the following days to discuss with us if you can (more details on the chats at the end of the post):
$ date -d '2026-04-21 15:00 UTC'
$ date -d '2026-04-28 23:00 UTC'
If you cannot attend these chats, and are already in personal contact with any SFC staffer, we all welcome you to chat with them or book a call — they would be happy to discuss. You can also reach out to <info@sfconservancy.org> and someone from our staff will engage in a slower email thread with you. As always, SFC never dictates mandatory policies on our member projects or our community. We look forward to hearing what you think at the sessions! My initial ideas follow:
Many people may recall Eternal September (in 1993) — when Usenet membership increased overwhelmingly — marking the annual September rush of student joins. The ensuing moderation challenges changed the culture of Usenet (the largest Internet discussion fora back then). Many early Usenet adopters left quickly. While this onslaught of “newbies” knew little of Usenet's traditional cultural norms, they nonetheless benefited greatly from these novel connections to discuss and learn together with people worldwide. The times were turbulent then, but eventually revised cultural norms emerged that benefited both the Eternal September arrivals and the old guard who stuck it out. Usenet (while less popular than it once was) survives today as the only widespread store-and-forward discussion fora optimized for low-bandwidth, spotty connectivity.
Today, another similar cultural change is afoot. I call it Eternal November. November 2025 — the month of Opus 4.5's release — has been widely identified as a similar inflection point: LLM-backed generative AI coding tools made a substantial leap to increased usefulness. A new wave of software developers — unfamiliar with traditional FOSS development culture — are adopting these systems quickly and making many (sometimes disastrous) mistakes. Some refuse to learn historical cultural norms, and others are genuinely confused (like all newbies) at the hostile responses.
We're all tempted — as some were in 1993 — to shun these “newbies” and reject their contributions aggressively. But what is really going on here? As in 1993, good and well-intentioned people just discovered newly interesting and hitherto unexplored ways to interact and engage with computing. FOSS's old guard reasonably feels angst, concern, and legitimate fear. We curated our norms over decades. But these norms are not sacrosanct. I urge us to reluctantly but seriously embrace this opportunity. We have much to teach these newcomers, and we must resist our arrogance of experience and assumptions that they've nothing to teach us.
I encourage all of us in the FOSS community to welcome the new software developers who've adopted these tools, investigate their motivations, and seriously consider cautiously and carefully discussing their workflows and consider if they may ultimately benefit FOSS. Seasoned software developers understand the benefits and limitations of LLM-assisted coding tools: we've studied for a few years the maintainability costs they incur, and the dangers of their flagrant, undisciplined use. We should give a try at explaining this to the newcomers. For example, if it comes from an individual (i.e., not a bot), an AI slop pull-request often offers an opportunity to educate a newcomer as to what they did wrong. Indeed, most FOSS developers that I know submitted a human-written-slop as their first contribution to a project. The ones who stayed are the ones who had an experienced developer explain what they did wrong — so they could improve and try again.
FOSS maintainers have decried for decades how few upstream contributions they receive. Maintainers today mostly bemoan the “AI slop”, and I agree that the slop is (at its absolute best) raw material to help make an actual contribution and (at its worst) tantamount to spam. Nevertheless, we must seek the goal of engagement to increase human collaboration (and thus maintainability). It is our job to recognize these opportunities, to communicate them, and to help new software developers build maintainable projects with us. It's a tall order: since the inception of FOSS, we've always needed to perpetually service the airplane while flying it: maintain the projects we have, while also creating the FOSS communities of the future for those projects. Only this approach has ensured FOSS will thrive and flourish. This is no time to give up that time-honored and documented successful approach.
The tiresome onslaught of slop (at least that presented by real humans) stems from a fundamental newbie confusion — but we can help newbies understand! We've always accepted (and encouraged) folks to keep private forks if they really like a patch that upstream determined was slop. My friend and colleague Bradley told me years ago that he maintained a personal fork of rsync for most of the 1990s because his extremely sloppy first-of-his-life submitted patch was rejected by upstream. Users who make a sloppy change that “works for them” should never be shunned; instead, they should be educated to make their own Git repository and rebase off upstream to their heart's content.
This is not to say we shouldn't encourage people to improve their behavior so they can contribute to and join FOSS projects. However, we do need to recognize when someone is ready and interested in doing that (and willing to put in a bit of extra time to help others), and when they just want to do a thing on their own, and that's fine for them. In the latter cases, we have always simply asked them to please stop sending us what they've got.
By recognizing what new technologies can do well, and what they can't, and how this differs from the technologies of the past, we can not only adapt our communities to these changes, but also take advantage of the new wave of people who are excited about our craft. It took time, but we adapted to the Eternal September. We can learn from that experience and do it again. The first steps are to recognize rapid changes, consider how they can benefit society, and then patiently and open-mindedly bring our ideals and extensive knowledge to the table to accelerate that positive change — while politely warning of the drawbacks (which are obvious to us but not to the newcomers).
There are many issues with LLM-backed generative AI coding tools; this blog post puts forward a few ideas about one of dozens of issues we now face (or will face in the near future). But every hope of improvement must begin somewhere. Therefore, if you help lead a FOSS project, or if you are using LLM-backed generative AI coding agents and want to learn to help rather than hinder upstream, please join us to discuss your concerns and the need to help welcome these new people to your communities in a manner that is productive. We will be running a series of interactive video chats, starting on April 21 at 15:00 UTC and April 28 at 23:00 UTC, in this room. SFC staff and volunteers will discuss these issues with the public. We'll consider how we can adapt FOSS projects to improve pro-AI contributor onboarding and how to better understand how newcomers are using and making software these days. Please also follow us using the linksa at the bottom of the page to learn about future such chats. SFC has dedicated 20 years already to successfully making FOSS the best it can be. We plan to do the same for the next 20 years, by understanding the needs of billions of new software developers, and welcoming them to our community with grace and education.
by on April 2, 2026
Last week, the Federal Communications Commission in the United States (the FCC) banned the sale of all new models of home routers not made in the U.S., which is ... all of them. The stated reason for this is that routers "pose an unacceptable risk to the national security of the U.S. or the safety and security of U.S. persons." A router manufacturer can apply for a "Conditional Approval" exemption to try and convince U.S. government bodies that their router should be allowed into the U.S., but this requires "A detailed, time-bound plan to establish or expand manufacturing in the United States" and "A description of committed and planned capital expenditures, financing, or other investments dedicated to U.S.-based manufacturing and assembly", and "an update on the status of their onshoring plan once a quarter" among other impractical asks. Devices built in the U.S. generally cost at least twice as much as devices built in Asia (see the Librem 5 (USA) for example) because U.S. manufacturing facilities are not ready with the scale and efficiency required to enable competitive pricing. The reason we chose to build the OpenWrt One in Asia is that it makes sure the device is as feasible as possible for people around the world to purchase. We expect it will take decades before the U.S. is ready to produce competitively-priced devices - user freedom can't wait that long.
And, in case you were hoping to buy an OpenWrt One, don't worry: the One has already received FCC approval so there is no change to its availability in the U.S. Naturally, we are concerned about the effect this has on any new hardware that SFC might develop, but this decision by the FCC does not create any near-term problems for us, or for FOSS generally.
We do applaud the FCC for recognizing how important home routers are to people's security. While the rulemaking is misguided, it's absolutely correct that the proprietary router manufacturers be accountable in relation to the hardware and software that individuals bring into their homes and their lives. We believe that manufacturers of routers that are primarily FOSS are in a much better position to evaluate the security of their devices, and so we analyzed the rulemaking taking into specific account its software aspects.
While the FCC decision focuses mainly on hardware, there are also some requirements for software. In particular, the FCC has hinted that it may restrict updates to existing hardware, in particular that existing routers "may continue to receive software and firmware updates that mitigate harm to U.S. consumers at least until March 1, 2027".
Since software updates to already-FCC-approved devices do not require a new FCC approval, it appears the FCC is trying to move beyond its usual authorization procedures to restrict what manufacturers are allowed to push to existing routers. However, the FCC notably does not restrict software changes made by owners of routers in the U.S. In particular, there is no indication that updates people make to their own routers, using software they have sourced themselves, would run afoul of any past or present FCC rule.
As a result, we do not believe that this new FCC decision affects whether and how people can run OpenWrt or other user-selected firmware updates on routers they have already purchased. Not only is this an important right in relation to our ownership and control of our own devices, it also ensures that people can keep their routers secure for far longer than the manufacturer may choose to provide security updates, by allowing them to install up-to-date community software that supports routers for 10, 15, or even more years after their initial release date, as OpenWrt does for many devices.
This leads us back to the stated goal of the FCC in making these changes: to ensure that routers do not "pose an unacceptable risk to ... the safety and security of U.S. persons." We certainly agree that all persons (including U.S. persons) should use technology that is safe and secure. And there are standards that exist to ensure this is the case, such as NIST IR 8425A, which the U.S. government already paid to research and produce and, alongside NIST, is recommended by Consumer Reports and other right-to-repair groups already. We have been assessing our existing processes (for OpenWrt, and especially the OpenWrt One) against NIST IR 8425A, and are now accelerating those efforts to ensure we can show that routers using OpenWrt are indeed safe and secure, as determined by independent bodies. This not only helps U.S. persons, but everyone around the world, as OpenWrt is available to anyone regardless of whether they are in the U.S. or not. We strongly encourage any regulation targeting safety and security to take a holistic view, recognizing that safety and security in our technology does not depend on what country we are in, but rather on common properties of the hardware and software we use, and a shared understanding of what technological safety and security means for all humans.
We have reached out to the FCC for clarity on this topic, and look forward to updating this post with their reply.
by on March 4, 2026
Earlier this week1, the U.S. Supreme Court (SCOTUS) denied certiorari (cert) in Thaler v. Perlmutter. Thaler contended that an image — generated by a Large Language Model (LLM)-backed Artificial Intelligence (AI) — deserved copyright registration. Since the U.S. Copyright Office refused to grant the registration, Thaler appealed to the U.S. District Court for the District of Columbia (DC Circuit). That Court affirmed the Copyright Office's decision. SCOTUS' denial of “cert” means they will not hear the case. Strictly speaking, this denial does not affirm the DC Circuit Court's ruling, but it does mean the DC Circuit decision stands.
Many in the Free and Open Source Software (FOSS) community raised concerns about the impact on copyleft — and even FOSS in general. TL;DR: Don't Panic! — this case is extremely limited in scope.
First, a proviso: this case is about copyright of an artistic image, not software. Copyright law — and the legal precedents around it — differ widely for different types of creative works. Analysis of the copyrightability of works of software varies in notable ways. Therefore, do not to assume that analysis for images apply broadly to software.
Second, while the decision is “published” 2, there are also many other cases related to LLMs and AI currently pending throughout the U.S. Courts. Courts and laws always lag behind technological advancement. Indeed, this is precisely why copyleft was invented: as a mechanism to achieve with existing laws and precedents what we could not accomplish in the legislature. Forty-one years after copyleft's invention, we still do not have a federal law that mandates software right to repair!
Third, the Court found that a registration was not valid (at this time) if the work's sole author is a computer program. Thaler (who was both (one of) the author(s) of that computer program and its user) repeatedly waived any claim to consider Thaler's own copyright in the LLM-backed AI prompting process. Thaler also did not argue any copyright interest in the LLM-backed AI system itself were subject of the registration. So, this decision does not evaluate any creative expression by (a) the author(s) of the prompts themselves, (b) copyrights held in the LLM, its weights, generation, curation, or its user interface, and (c) copyrights held in underlying works in the LLM training data.
Thaler's original registration was the root cause of this substantial narrowing because the registration contended that the AI system itself was the author of the image. This case only considers a copyright registration where the sole “author” is identified as a specific computer program. Thaler stipulated that the work was generated solely through prompts and no human modified the work thereafter. As such, even if the other districts begin citing this case regularly, and even if many districts decide it applies to software without further consideration of the difference in the types of works, such precedent causes no disaster for FOSS.
Admittedly, some LLM-backed generative AI agents can be merely prompted to create a work of software from scratch that has some transient utility. However, the most common workflow in using these agents (at least in FOSS development) is as follows:
Furthermore, dicta 3 — appearing in the DC Circuit ruling — supports a conclusion that the human actions on that third step would constitute creative expression — affixed in tangible medium — suitable for copyright registration. Indeed, their ruling states:
First, the human authorship requirement does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence. The rule requires only that the author of that work be a human being — the person who created, operated, or used artificial intelligence — and not the machine itself. — (Thaler v. Perlmutter, 130 F.4ᵗʰ 1039, 1049 (D.C. Cir. 2025))
The Court also indicated these other issues are for a future time in another case. The DC Circuit readily admits that their ruling applies only to the state of AI systems at the time of writing4. Again quoting from their ruling:
Of course, the [Thaler's AI] Machine does not represent the limits of human technical ingenuity when it comes to artificial intelligence. Humans at some point might produce creative non–humans … Science fiction is replete with examples of creative machines that far exceed the capacities of current generative artificial intelligence. For example, Star Trek’s Data might be worse than ChatGPT at writing poetry, but Data's intelligence is comparable to that of a human being. See Star Trek: The Next Generation: “Schism” (Paramount television broadcast Oct. 19, 1992) (“Felis catus is your taxonomic nomenclature, an endothermic quadruped, carnivorous by nature”). There will be time enough for Congress and the Copyright Office to tackle those issues when they arise. — (Thaler, 130 F.4ᵗʰ at 1050)
[ As always, SFC is not a law firm, IANAL, and TINLA. ]
1 I — and my colleagues at SFC — acknowledge that SCOTUS made other decisions recently regarding an array of important social justice causes. Since SCOTUS' decision to deny cert in this particular case is so closely related to my work, I'm writing about it. However, all of us at SFC acknowledge that our community is reeling from other recent decisions.
2 In this context, “published” is a term of art that lawyers use to describe a case that the publishing Court (in this case, the DC Circuit) felt was important enough to “share officially and formally” with other Courts. While (in a precedent-based legal system) any Court can cite an unpublished case from another Court, published cases are much more likely to be cited than unpublished ones.
3 “Dicta” is explanatory language found in a court's decision that isn't necessary to the court's conclusion. . Dicta isn't precedential but it can be persuasive.
4 Note that the DC Circuit issued their ruling in March 2025. It is not uncommon for SCOTUS to delay for a year (or more) before issuing a ruling to grant or deny cert.
[1] 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69