Every Go application since 1.24 will activate MPTCP on their sockets by default.
And I’ve read that this includes Syncthing, which is why I’m posting this here
This is shady as hell. It’s an AI company again (Anthropic), they do not follow rules of responsible disclosure (again), it’s a race condition again, they do not provide a proof of concept (obviously since they shouldn’t talk about it at all until after we got the patch) but instead create big media fuzz about their glorious AI… hell, they do not even tell us the kernel versions or anything.
If we find out eventually that you have to disable default mitigations for this to become dangerous (like the stunt they did with Firefox) it’s too late, the media won’t give a shit anymore. Like we just had with PHP - huge panic, enormous media circus, barely any real-life consequences in the end and the responsible disclosure process being a total mess to blow it up even more.
Honestly, every time I see some AI company hyping yet another huge and super dangerous exploit only their superior AI found “after X years of nobody seeing it” I now expect a nothingburger.
We’re probably fine, nothing out of the ordinary to see here.
The Php vulnerabilities were and are being exploited in the wild
For server crashes, but not for RCEs. The only proven case of this I know of was with a specifically prepared PHP instance as honeypot that had an optimal attack surface and was done by Perplexity (another AI company). Outside of that there’s no credible large-scale RCE exploitation that’s any worse than the bugs that get fixed on a daily basis. It doesn’t even remotely justify the panic and media frenzy.
it was merely dangerously exploited by AI bros.
I agree the media frenzy was overblown but it seems legit at this point.
Network devices were getting popped. NAS were also.
They found a previously unknown vulnerability and it was fixed:
“This vulnerability (CVE-2026-98432) was patched upstream and backported to all maintained stable kernel branches.”
This is good no?
Not saying fixed bugs aren’t good, I’m just so fed up with this marketing-driven fearmongering. They’re overhyping every god damn bug they can because they know most people won’t know how bad it actually is. They do worse security practices than most homelab enthusiasts and then prophesize the AIpocalypse because their linear algebra “escaped” (the door was wide open). Hell, they don’t even manage to understand proper bug disclosure practices. They’re playing the “Social” Media outrage game on a global scale to drive sales and everyone is suffering for it. I’m so, SO done with all of this bullshit. When the bubble finally crashes I’ll throw a raging party with blackjack and hookers.
They’re overhyping every god damn bug they can because they know most people won’t know how bad it actually is
Where is Anthropic doing this? The one that finds the vulnerability are not the ones that assigns it a CVSS score. CVSS is calculated with a standardized format, and is not mean to be a measure of risk. So this is no the AI companies saying how bad the vulnerabilities are, but rather the owners/maintainers of the code.
Hell, they don’t even manage to understand proper bug disclosure practices.
Have you ever stopped to consider previous bug disclosure practices aren’t as applicable? If anyone can find bugs using AI, then even if Anthropic or OpenAI finds a vulnerability, they can’t know for sure they are the first. Even if they are the first, it’s only a matter of time before the next people find it even if the first discovery was never publicly disclosed.
There have also been countless bugs given directly to companies, and fixed without a big blow up or news spectacle. You didn’t hear about those, because they didn’t blow up. So you are going to have a level of confirmation bias.
Why couldn’t it be that there are just a ton of really bad vulnerabilities that have never been discovered until a tool like AI came by to find them? There is nothing special about Anthropic or OpenAI, they are just the fastest moving with this new technology. Why couldn’t the bugs they find be just as bad as they are claimed to be, especially when severity is often not determined by the AI company anyway
Could everyone stop blowing holes in things please, spent all yesterday rebuilding internet facing servers.
On the plus side, everything will be up to date. So there’s that
Also with both this kernel and the php thing, Debian had a fix way before the big corporations. Even Amazon
that’s the point. you’re supposed to be exhausted enough so that you just relent and let AI take over your duties.
second objective is to showcase your incompetence to leadership so they are scared enough to replace you with AI.
this is the plan.
rebuilding internet facing servers
Because of one vulnerability?
Jebus. Response at one site was a 10-minute thing and then letting cron do the work.
Where the hell is your D-R? I worry you’ve made some very non-optimal decisions in the past and are taking on way too much work now.
It took longer than planned as we are in the process of getting everything sorted for quick rebuilds, most of the time was waiting for GitHub to not poo it’s self.
If we had everything in ansible it would be 5 minutes
If we had everything in ansible it would be 5 minutes
Ansible at the day-job takes about 45 min to do a patch. As someone who’s used a fair number of alternatives, maybe consider anything else on the planet?
But, seriously. The response to this particular sploit wasn’t really “config management”-y. It was a simple process that cron generally handled anyway, and network segregation and firewalling allowed us to prioritize easily. My worry is that there are some core architectural decisions you may need to revisit in the post-mortem on this, because there may be some fundamental changes required to get sensible wins in the future.
(If you’re on containers, I weep for your validation challenges; also, get bazel)
We find ansible to be the best way to manage our fleet, patching and keeping config in sync.
That’s not to say it can’t be slow sometimes, but it works for us.
Does unattended upgrades fix this on Debian? Or do I have to reboot now for it to take effect?
Depends on your settings. You need to reboot for kernel upgrades, so you’re good if you have automatic periodic reboots configured and kernel updates applied unattended.
Is it known if/ when Ubuntu is going to provide a fix? https://ubuntu.com/security/CVE-2026-98432 Rn the page states: work in progress for a fix.
Someone rated the issue as 9.8.Other distributions provided a fixed kernel package 10days ago. Some say, Ubuntu engineers do not rate the issue as critical. But there are exploits available which prove a DoS. Not sure if the RCE with privilege escalation has an exploit yet and under which circumstances the RCE works. Is a golang service required? Are other services affected - is multipath TCP sufficient? Does the issue affect ipv6 too?
DoS isn’t that bad. RCE with root + an exploit on the other hand. Um…
if you run Ubuntu as a server nobody should take you seriously.
Debian literally exists as a server first while Ubuntu exists as desktop first.
the only valid excuse to using Ubuntu server is if your hardware is a RPI or rpi-clone and at that point you’re at a hobbyist level which makes more sense why you’d be using Ubuntu at all.
if you run Ubuntu as a server nobody should take you seriously.
May I ask: Why do you recommend to not use Ubuntu as a server OS?
They offer a server version, provide LTS releases and there is a (limited) extended security maintenance program if one decides to use a distribution after the EOL of the LTS. Which led me to believe it was a viable option.
I’m seriously interested in your insights.
Ubuntu is notoriously finicky as a server. multiple issues with how system configuration is different from other servers like Debian, redhat, etc. firewall used to be an absolute terror. also, as described, your patch levels are tailored to the whims of a corporate entity that historically has sided with desktop stability and experience over server stability and performance.
while we’re at it, canonical captures and sells system information like Microsoft does. by using Ubuntu, you become the product. IDK how much you know about companies that sell their consumers data, but typically they put profits over security.
I’ve worked for a handful of MSP/CSP vendors over the years and they only offer Ubuntu as a server option with one big caveat. you are responsible for the maintenance of the system. lo and behold, four years later they always come crying about how they’re about to fail an audit because it’s never been updated, or is misconfigured, or was hacked because they installed a bad library, ad nauseam.
from my own experience using it, though aged, it came with a service that was listening on port 3306. was it mysql? nope. what was it? dunno. I killed everything on that system that wasn’t required by the kernel. still listening. couldn’t install a database on it, so it was wiped and I never used it again.
simply put, Ubuntu is a desktop first OS. if you use it as a server you’re going to have a bad time because statistically speaking only amateurs use it that way.
From what I read about regarding the PHP vulnerabilities, the RCE was extremely difficult to pull off. I’d guess it’s the same here, at least until some non-AI companies weigh in, since these AI companies tend to distort the reporting and severities quite a bit.
Thank you.
Difficult as in unlikely that the exploit works or difficult to write an exploit? The latter doesn’t help me much. :)
Regarding the php2shell issue: a lot of admins took their services down until it was patched. It’s more a decision based on a guess - big number bad, evil acronym scary - than actually understanding what the exploit does. One can either believe it or not.
And that’s the thing about the recent exploit (multipath TCP): I can’t evaluate myself, whether the issue is likely an exploitable RCE. That is one factor missing in risk assessment: is the outcome potentially fatal? Yes, rce and priv escalation. Is it likely? I can’t tell.
I’m sorry. I don’t mean to doubt your expertise. I simply don’t know you. It is the same with hackernews, the author repeats what he’s or she’s told. (Good overview article though. Thanks again for it.) RN It’s: Someone from the internet said don’t panic. A link to a statement from the Ubuntu package maintainer or distribution would make me feel at ease. I still couldn’t make sure for myself - a lack of time and competence in the field - but I would trust it.
Difficult as in unlikely that the exploit works or difficult to write an exploit?
Difficult as in it’s a race condition that can theoretically be exploited when you get the timing right. But timing, if it’s not on a well-known isolated system right before you in some lab, depends on a million details, some –once we are talking about real-life examples– outside your control.
Yes. Anthropic wrote: it took them like an hour to get it right on their lab system. That’s different than the wild wild internet.
