> and since nobody stepped up to help us deal with the influx of the AI-generated bug reports we need to move it out of tree to protect our sanity.
This thread from the linux-hams mailing list [2] has more insight into this decision. I guess the silver lining is that, more modern protocols (in userspace), written in modern languages will become the norm for HAM radio on linux now.
[1] : <https://lwn.net/ml/all/20260421021824.1293976-1-kuba@kernel....>
[2] : <https://lore.kernel.org/linux-hams/CAEoi9W5su6bssb9hELQkfAs7...>
That's really it. The list of things that "need" to be in the kernel is shrinking steadily, and the downsides of having C code running in elevated privilege levels are increasing. None of that is about LLMs at all, except to the extent that it's a notable inflection point in a decades-scale curve.
The future, and we basically all agree, puts complexities like protocol handling and state in daemons and leaves only the hardware, process and I/O management in the kernel.
Basically, Tannenbaum was right about the design but wrong about the schedule and path to get there.
And this is very dependent on the hardware programming interface of the devices.
Look at AMD who is investigating hardware user-level queues for their GPUs (dunno if this is possible because of VMID stuff)
Be real with yourself, do you know anyone using ISA or PCI in 2026? Everything is built on PCI-E except in specific industrial settings or on ancient hardware that's only relevant for retrocomputing. Is anyone using the ATM network protocol anymore? MPLS and MetroE mostly replaced ATM, and now MPLS is being largely supplanted by SDWAN technologies and normal Internet connections. I have been doing networking nearly my entire career in some capacity, the last time I touched X.25 or Frame Relay was in the early 2000s, the last time I touched ATM was in the mid early 2000s... the last time I touched ISDN was in the mid 2010s, and that was an IDSL setup, which is itself a dead technology. The last laptop I owned that had a PCMCIA card slot was manufactured in 2008.
I don't want to see these capabilities completely disappear, but there's no reason they should ship in the mainline kernel in 2026. They should be separated kernel modules in their own tree.
The main issue with this is that by being on a separate tree they do not benefit from the API breakage updates in the kernel. After all the main benefit that kernel devs mentioned over the years for keeping drivers in the kernel instead of separate trees is that the code gets updated whenever internal APIs change.
The only problem here, if any, is the false sense of confidence given by LLMs to people who have no business touching kernel code.
Stuxnet should have been a wakeup call to everyone: the boring, obsolete, “safe because nobody browses TikTok on it” hardware is exactly the highest risk.
Now those are niche use cases, but they do exist. However, what’s wrong with removing insecure code for these niche cases? Either someone will step up to actually maintain it, or newer versions of the kernel will be leaner and have less historical cruft.
Xbox/PS controllers, for example. I believe some old RAID controller and WiFi drivers are removed too. Whatever they don't want to support.
Generalizations that lose accuracy are not valid. “Ice cream is sweet, and candy is sweet, so food is sweet” is reductive.
Software exists to be used, not to be secure. These are not useless pieces of code. If they were useless, then no one is using them, so there is no security risk. This is equivalent to turning off (or destroying) a computer to secure it.
Alternatively (and I'm disappointed Linux/Greg K.H. haven't done this), drivers and other isolated modular code should be marked as unmaintained, and for those with reported vulnerabilities, a similar config flag set. Require explicit acknowledgement by kernel builders to include them in the build config.
Things have been trending badly with Linux in this area, it feels like it's lost it's original calling, and is now heavily influenced by PR and corporate interests. The desktop Linus used in the 90's to write Linux should be able to run the current Linux kernel. But it doesn't even support the CPU architecture any more!
Some of us have perfectly good old hardware we can put to modern (non-networked) use, but we have to either use netbsd (if it supports the task/program), or generate more e-waste and dump the hardware in the bin. And buy yet another RPI, and waste money and resources. But at least, so long as it is simple use cases that don't require modern software, you can just slap an old version of Linux on it, but at least in my experience, stability was more of an issue for older drivers than it is today, so Windows 98 or XP is a better choice sometimes for x86.
why do they even need to be in kernel repo and not brought at/after install time?
I wrote and maintained 10GbE drivers for a small company in the 2000s, and just the SHIM file for our driver on Linux to massage over API differences was well over 1000 lines. I think it was close to the same size as the entire driver for one of the BSDs.
It's not a nightmare anymore to port drivers
Not that it's not going to happen; Linus is already a vibe-coder, and several maintainers have fallen for the LLM crap as well.
Yes, I don't see the point of maintaining technical debt just for the sake of it.
The security environment in 2026 is such that legacy unmaintained code is a very real security risk for obscure zero-days to exploit to gain a foot in the door.
Reading through the list I don't see it being an issue for the overwhelming majority of Linux users.
Who, for example, still uses ISDN in 2026 ? Most telcos have stopped all new sales and existing ISDN circuits will be forcefully disconnected within 3–5 years as the telcos complete their FTTP build-outs and the copper network is subsequently decomissioned.
Most TV and radio stations.
- Nobody is familiar with the code
- Almost all of the recent fixes are from static analysis
- Nobody is even sure if anyone uses the code
This feels a lot like CPython culling stdlib modules and making them pypi packages. The people who rely on those things have a little bit of extra work if they want a recent kernel version, and everyone else benefits (directly or indirectly) by way of there being less stuff that needs attention.
The overlap of bugs being found, nobody caring enough to bother read the reports or fix the code, and nobody caring that the modules are pushed out of main seems good.
In general, drivers make up the largest attack surface in the kernel and many of them are just along for the ride rather than being actively maintained and reviewed by researchers.
Can't wait to AI braindead folks get collapsed down for the good.
At the same time, a lot of these bugs were in places that people weren't looking because it's not actually important. This kernel code had already been a longstanding problem in terms of low-effort bot-driven security reports and nobody had any interest in maintaining it. So this was more LLM-assisted technical management than LLM-assisted security, it finally made a situation uncomfortable enough for the team to do something about it.
Another example: Mythos found a real bug in FreeBSD that occurs when running as an NFS with a public connection. But... who on earth is doing that? I would guess 99.9% of FreeBSD NFS installations are on home LANs. More importantly, Anthropic spent $20,000 to find this bug. Just think in terms of paying a full-time FreeBSD dev for a month and that's what they find: I'd say "ok, looks like FreeBSD has a pretty secure codebase, let's fix that stupid bug, stop wasting our money, and get you on a more exciting project."
I do think anyone who has a legacy open-source C/C++ codebase owes it to their users to run it by Claude/Codex, check your pointers and arrays, make sure everything looks ok. I just wish people were able to discuss it in proper context about other native debugging tools!
In terms of quality ("are there bugs that professional humans can't see at any budget but LLMs can?") - it's not very clear, because Opus is still worse than a human specialist, but Mythos might be comparable. We'll just have to wait and see what results Project Glasswing gets.
Either way, cybersecurity is going to get real weird real soon, because even slightly-dumb models can have a large effect if they are cheap and fast enough.
EDIT: Mozilla thinks "no" to the second question, by the way: "Encouragingly, we also haven’t seen any bugs that couldn’t have been found by an elite human researcher.", when talking about the 271 vulnerabilities recently found by Mythos. https://blog.mozilla.org/en/firefox/ai-security-zero-day-vul...
The plural of "Opus" is "Opera". Might be a tad confusing tho :)
Being flooded with these kind of reports can make the actual real problems harder to see.
IMHO, if you were to do a manual audit of the Linux kernel, the first thing to do is exclude all the stuff you're never going to run, because why spend time on it?
These scans are looking at everything, because once you set it up, the incremental cost to look at everything is not so bad.
This is going to push lesser used stuff out of the mainline, which sucks for people who were using it, but is better for everyone else.
Of course some people don't do that, and send all the reports anyway... and then scream from the hilltops about how incredible LLMs are when by sheer luck one happens to be right. Not only is that blatant p-hacking, it's incredibly antisocial.
It's disingenuous marketing speak to say LLMs are "finding" any security holes at all: they find a thousand hypotheticals of which one or two might be real. A broken clock is right twice a day.
Modern LLMs with a reasonable prompt and some form of test harness are, in my experience, excellent at taking a big list of potential vulnerabilities and figuring out which ones might be real. They're also pretty good, depending on the class of vuln and the guardrails in the model, at developing a known-reachable vulnerability into real exploit tooling, which is also a big win. This does require the _slightest_ bit of work (ie - don't prompt the LLM with "find possible use after free issues in this code," or it will give you a lot of slop; prompt the LLM with "determine whether the memory safety issues in this file could present a security risk" and you get somewhere), but not some kind of elaborate setup or prompt hacking, just a little common sense.
What commenters don't seem to understand is that especially CVE spam / bug bounty type vulnerability research has always been an exercise in sifting through useless findings and hallucinations, and LLMs, used well, are great at reducing this burden.
Previously, a lot of "baseline" / bottom tier research consisted of "run fuzzers or pentest tools against a product; if you're a bottom feeder just stuff these vulns all into the submission box, if you're more legit, tediously try to figure out which ones are reachable." LLMs with a test harness do an _amazing_ job at reducing this tedium; in the memory safety space "read across 50 files to figure out if this UAF might be reachable" or in the web space, "follow this unsanitized string variable to see if it can be accessed by the user" are tasks that LLMs with a harness are awesome. The current models are also about 50% there at "make a chain for this CVE," depending on the shape of the CVE (they usually get close given a good test harness).
It seems that the concern with the unreleased models is pretty much that this has advanced once again from where it is today (where you need smart prompting and a good harness) to the LLM giving you exploit chains in exchange for "giv 0day pl0x," and based on my experience, while this has got an element of puffery and classic capitalist goofiness to it ("the model is SO DANGEROUS only our RICHEST CUSTOMERS can have it!"), I believe this is just a small incremental step and entirely believable.
To summarize: "more efficient than all but the best" comes with too many qualifiers, but "are LLMs meaningfully useful in exercising vulnerabilities in OS kernel code," or "is it possible to accelerate vulnerability research and development with LLMs" - 100% absolutely.
And you don't have to believe one random professional (me); this opinion is fairly widespread across the community:
https://sockpuppet.org/blog/2026/03/30/vulnerability-researc...
https://lwn.net/Articles/1065620/
etc.
No.
Like everything else an LLM touches, it is prone to slop and hallucinations.
You still need someone who knows what they are doing to review (and preferably manually validate) the findings.
What all this recent hype carefully glosses over is the volume of false-positives. I guarantee you it is > 0 and most likely a fairly large number.
And like most things LLM, the bigger the codebase the more likely the false-positives due to self-imposed context window constraints.
Its all very well these blog posts saying "LLM found this serious bug in Firefox", well yeah but that's only because the security analyst filtered out all the junk (and knew what to ask the LLM in the prompt in the first place).
Another way to see this is that you mentioned "LLM found this serious bug in Firefox", but the actual number in that Mozilla report [2] was 14 high-severity bugs, and 90 minor ones. However you look at it, it's an impressive result for a security audit, and I dount that the Antropic team had to manually filter out hundreds-to-thousands of false-positives to produce it.
They did have to manually write minimal exploits for each bug, because Opus was bad at it[3]. This is a problem that Mythos doesn't have. With access to Mythos, to repeat the same audit, you'd likely just need to make the model itself write all the exploits, which incidentally would also filter out a lot of the false positives. I think the hype is mostly justified.
[1] https://lwn.net/Articles/1065620/
[2] https://blog.mozilla.org/en/firefox/hardening-firefox-anthro...
[0] not trivially if you want to validate if it works
Of course. But that's true for all userspace solutions too, and there are many options for async APIs (io_uring et. al.) which work to address that.
The point is that you want the IP stack (or whatever) to be passing stuff around on unix domain sockets for every packet. It's that you want it running in its own memory domain.
Yes, what we see coming out of the bottom of funnel is now is a little better. But it's sort of like reading day trading blogs: nobody shares their negative results, which in my direct experience are so bad they almost negate any investigative benefit. I also think part of this is that a small set of very prolific spammers were sufficiently discouraged to stop.
> "Remarkably few of them complete false positives."
To be clear, I'm not saying 0% false-positive because that will always be impossible with any LLM.
However, to greatly over-simplify what I already said ...
The presence of >0 false-positives means you still need someone who knows what they are doing behind the keyboard.
The presence of an LLM, no matter how good, will never remove the need for a human with domain expertise in security analysis.
You cannot blindly fix stuff just because the LLM says it needs fixing.
You cannot report stuff just because the LLM says it needs reporting.
There may well be scope for LLM-assisted workflows, but WHO is being assisted is a critical part of the equation.
That is the fundamental point I am making.
> You cannot report stuff just because the LLM says it needs reporting.
Not today, maybe. Though with a good enough harness we're pretty close to seeing that already.
But in 6 months after another halving of the error rate? I wouldn't be so sure.
Opera is the traditional plural from Latin, now perhaps for more scholarly use in English.
Results from a quick search.
I doubt it. And as I said, telcos have ceased new sales of ISDN and will be shutting down copper networks within 3–5 years.
Therefore if there are still TV and radio stations still using it, they will be forced to stop using it by circumstance, i.e. they will find their ISDN will cease working after the telco shuts down the kit in the exchange.
Losing support in software however, does.
Since you claim to be a domain expert, give me hard named examples with independently verifiable links. At this stage I want facts, not anecdotes.
Because right now, my semi-educated guess is they are all using IP-based streaming codecs and protocols for remote contributions, outside broadcast, studio links and pretty much everything else under the sun.
> As part of our continued collaboration with Anthropic, we had the opportunity to apply an early version of Claude Mythos Preview to Firefox. This week’s release of Firefox 150 includes fixes for 271 vulnerabilities identified during this initial evaluation.
We're in a new era for security. You're either using AI to catch vulnerabilities in your code... or someone else is, and 0wning you.
You can do a neat trick with this if you set up IP over AX.25, particularly with softmodems. Since you've got IP you can do SSH or TLS over it, right? At least, if you set all the timeouts really long, because some of those packets take a while at 120 bytes per second.
So then you can tune the tones to be a little off the normal frequencies of one side, and play them through speakers with two PCs connected together. When you ssh from one to the other, you will hear the establishment packets and the flurry of packets for every keypress pingponging backwards and forwards between the two systems.
Absolutely brilliant for demonstrating how things like TCP works with retries (plug a mike into it too, shout some interference) and how UDP doesn't, and stuff.
ok, what are the niggly details. we don't have interrupts, and we're running under the general scheduler, so there may be some effects from not getting scheduled as aggressively.
we still need to coordinate through the kernel wrt port bindings, since those are global across the machine, but that's just a tiny bit.
clearly we may be re-opening a door to syn-flooding, since the path to rejection is maybe longer. maybe not, but maybe we can leave the 3-way handshake in the kernel and put the datapath in userspace.
we probably lose rtt-estimates hanging off of routes.
none of that suggests 'several times slower'
Seemingly we've been writing kernels for years, and they still are full of security holes.
Do you realize AX25 it's just something loaded on demand when the user requires it, and not by default? Do you know the basics on how the systems work bellow your shiny UI's and IDE's?
First, AX25 modules would just lie down in the disk harmless, no AX25 stuff it's loaded unless some user modprobe thems in order to setup some hamradio stack with HamNet and the like.
I see far more security issues with blobs loaded in a so-called GPLv2 kernel everywhere where the tarball almost weights more in blobs than in libre source code. Yet these LLM bootlickers will happily accept whichever non-free firmware on their noses.
Somehow propietary Radeon, Nvidia, some Intel audio drivers for SOCs and the tons of ARM related firmware blobs are not a security issue. At all. Just kick random bits over the BUS without knowing what really happens with the device. Even if some of them can have full access to the RAM and CPU and the like. That's pretty fine. Ah, yes, IOMMU's and the like. Not enough for some cases. Sorry, but these people can't be serious where the actual multi-CPU based networked computer it's full of opaque bits where you have no control on what they do at all.
For clarity: the example upthread about pwning was TCP/IP, not AX.25.
Also the idea that "there are no local exploits in this kernel code because it's not used by the running system" is like the proximate cause of 80% of local privilege escalation vulnerabilities. Seriously?
I think it's also a fully switched system so you are guaranteed bandwidth with no packet drops or buffering which is clearly useful for broadcast work.
https://en.wikipedia.org/wiki/DOCSIS
https://defcon.org/images/defcon-16/dc16-presentations/defco...
But how much is this to do with either Channel 4 not supporting (in the "assistance" sense of the word, rather than "interop") his move to IP or potentially his personal reluctance to change ("ain't broke don't fix it" mindset).
Given he is in the UK and the incumbent telco (BT) are switching off ISDN in 2027, I really suspect there is more than meets the eye to your friend's story.
I am not seeking to judge, I just feel realistically that it highly unlikely that at this late stage (1 year to go to 2027) there really is no other option other than ISDN when collaborating with Channel 4...
The reason I say this is because even the briefest of internet search throws up hard publicly-available evidence that the broadcast world have indeed moved on in the world ....
Way back in 2008 the BBC were already investigating options to move away from ISDN...[1]
... and evidence is out there the BBC are using SIP for critical things like remote Radio contributions[2]
> guaranteed bandwidth with no packet drops or buffering
You can absolutely get this on IP.
[1] https://downloads.bbc.co.uk/rd/pubs/whp/whp-pdf-files/WHP170... [2] https://support.inquality.com/kb/faq.php?id=144
If you have that nickname you would already know that, for sure.
So the TL;DR is we are not disagreeing then ? ;)
I never expressed any doubt that traditionally ISDN absolutely was the lifeblood of broadcast, there is zero doubt about that.
What I am saying is that was then and now is now. We are now sitting here in 2026 and the world of comms has moved on dramatically and the broadcast world has moved along with it and that those people still clinging on to legacy ISDN will be forced to shift to IP-based technologies because they will be forcibly disconnected by their telcos very soon (1–5 years).
The reality is also that here in 2026 we live in a world where (a) you have a 4k tv and high-end audio system in your home ... so there is a natural limitation on what utility ISDN has in this world, and (b) the general public is increasingly consuming the media produced by the broacaster via IP means (streams over 5G-IP to mobile, streams over IP to Apple TV boxes) ... so if a broadcaster can escape the un-necessary complexity (and cost) of transcoding ISDN-received content to IP and shift to an IP-to-IP environment, why would they not want to do that ?
- Use Mosh instead of SSH
- Spawn TMUX in the remote machine to send less bits per session
I tried Mosh+Tmux with 2.7 KBPS (and less) when I was using a data plan. It worked perfectly fine, no delay or barely noticeable.
I used to use mosh and tmux over 9600bps AX.25 before I had 3G data, a very long time ago. Strictly speaking SSH over amateur radio breaks the rule about encryption but 144MHz is a big place with no-one in it, and you can't pay Ofcom to take an interest in what people do on amateur radio.
This kind of chicanery is the vanilla pudding of the hacker world. It's everywhere. Suffice it to say that you're simply wrong: NO, it's never OK to argue a subsystem is safe because you personally think it can't be loaded. It 100% can be, that's the easy part.
Not an issue for AX25 per se.
If you can fool an user to run root instructions, it's game over, period.