So it begins. The war against the abliterated AI models. Even at 1,000 TPS any service is a target.
Imagine abliterated frontier AI models reach >20,000 TPS. As I said before [0] you are going to have to hire specialized technical bodyguards from the senior to senior staff level using AI agents to protect your entire business.
Primeagen went over this in the last episode or two - apparently this sort of defense is costing ... one of the big models or companies hosting them ... $900k/day or something to constantly run the best model and monitor 24/7 to combat the attacks [0]
I don’t really have experience running anything like this so this is very much a comment front the peanut gallery but
Is that really the best way to do things? I’m not really sure what an agent offers here does it just look at logs and do actions? I would not really feel safe if that’s my service’s answer to security.
I could see it maybe as a really smart filter to block ips that are sending probing requests but I don’t really see that as a good reason to have an agent be sitting 24/7 looking at logs.
OpenClaw agents continuously scanning your prod infrastructure and applying config tweaks and package updates autonomously on the fly in response to viral CVE numbers trending on X.
"You got it. Corporate core data for Tessier-Ashpool S.A., and that ice is
generated by their two friendly AI's. On par with anything in the military
sector, looks to me. That's king hell ice, Case, black as the grave and slick
as glass. Fry your brain soon as look at you. We get any closer now, it'll
have tracers up our ass and out both ears, be tellin' the boys in the T-A
boardroom the size of your shoes and how long your dick is."
I recommend Gibson’s (abridged) reading of the book - but only after reading the book. Whenever I read a quote from the book now I hear it in Gibson’s iconic voice.
Unikernels (e.g. MirageOS) were gaining traction right before containers swooped in and sucked the air out of the room. I think they could see renewed interest in the very near future.
More likely we'll just see another shakeup of the layers of abstraction that the average company operates at. Why would you hire specialized people when AWS can do so for you?
You are being downvoted but I agree, any company including openAI, anthropic, google etc... should be closely regulated by goverment to ensure they never run any AI that doesn't follow all the rules defined by governing bodies, nor train them to have any workaround... NO ONE HAS ABLITERATED ACCESS TO ANYTHING.
Thanks god, you are not a wacko thinking qwen3.0-1.5B-abliterated would hack a goverment.
Wow, that isn't clear.
Is there an incident, or did they just become aware about MPTCP vuln?
Details on the mptcp are also pretty scarce as well right?
It is almost certainly the latter, since it is another “Customer Security Advisory”. There is a currently open report for a tranche of nginx vulnerabilities[0] and a closed report for PwnEscape[1] and the PHP RCE[2] that are identical. The HN topic has been editorialised.
Based on email I got from another big hosting provider, they're all patching for the same checksum metadata vulnerability (the CVE they mentioned is the same). There is however a root RCE PoC according to the researcher so...potentially bad.
It feels like we hit an inflection point in the last week or so. Kind of a large tidal wave to crash on the coast, but likely not the tsunami yet. I hope.
If you are running services and don't yet have an easy way to react to these things with swift process defined, I hope you don't like to sleep.
When mythos was released - I guesstimated that we had 6 months until hackers had similar enough capabilities that we would start to see real problems in the wild. Unless there is some change in the trend line things are about to get really quite spectacular.
This isn't Linode getting attacked, it is telling customers that this zero click RCE CVE is pretty serious. You should patch ASAP.
Did you know that GO defaults to listening on TCP connections with MPTCP, so anyone running GO programs listening on the internet (Caddy, traffic, etc) and are on an unpatched kernel now have a way to from attackers to get root on that box?
Even if you're not using MPTCP, this functions as a DOS attack.
It's a very bad CVE, so yes. Follow the advisor and patch.
What if the "inflection point" is just all of the low-hanging fruit that can be detected by agentic vulnerability scanning? If the low-hanging CVEs get patched we might not see a tsunami.
And even patching the ‘small’ cves will take us for a spin as it already has. Then there can be specific low hanging fruits that chained together make a splash.
All companies should at this point invest a lot of money into defensive AI. In ~July the AI hit crypto and bitcoin, where it's the easiest to extract the most money.
Now the AI powered attacks are hitting infra and SaaS. Next (~December), it will hit classic infrastructure, banks, small businesses, etc.
By mid next year, we will be over this hump with defensive AI being required part of everything.
I find it breathlessly funny how people think it's going to be AIs vs AIs fighting it out with light sabers and plasma lasers in the cyber space or some shit.
Your intuition that this can be balanced out is wrong. When you have a CVE in your TCP stack, your vault unlocks and you're fucked, it's done. My dude, there is never going to be AIs scanning every incoming packet for signs of a 0day, and deciding whether to, uhh, cut it up with a lightsaber and shoot it with a laser pistol. Your options are to have a stack so secure that it can't be targeted, or to disconnect all network cables.
I specifically meant red teaming with AI across your infrastructure to find vulnerabilities before these are exploited. This is what I have been actively doing and it works.
Anyone have any thoughts on what happens when banks/brokerages start getting hit in earnest? Think it's prudent to split single brokerage portfolios across multiple? Or do they have enough speed bumps in their procedures that they're not quite so vulnerable to major losses?
If the large and politically connected banks and brokerages have losses large enough to threaten their survival as institutions then the government will jump in to help secure them and bail them out if needed.
I would stick with a "too big to fail" institution instead of splitting across multiple.
If you want a real "backup" you could ask for physical stock certificates or get physical gold or something else that cannot be taken from you by computer hacking.
FDIC insures deposits up to 250k. Big institutions basically all have this government backing. If an incident occurs, FDIC has deposits available next business day in event of a bank failure
FDIC insurance is mostly unrelated to this. If someone hacks a brokerage, sells a bunch of customer assets for cash, adds new bank accounts for those customers, and then withdraws that cash to those new bank accounts, FDIC insurance does not help at all.
I work in infosec and work with many of these large organizations.
Simply put I don't think dropping in smart people in dysfunctional organization is going to have instant results. It will take a pretty long period of time to make them aware of all the systems with holes.
In the meantime hackers would be running off with actual information and money. More than likely I see banks turning off all kinds of systems in a panic and people with their money there being in an even bigger panic.
Anybody recommend a CVE alert self-hosted app? E.G. I specify my software, OS’s, and versions and when a matching CVE is created I get an alert. Don’t want enterprise security products.
Why was the previous discussion of this flagged? https://news.ycombinator.com/item?id=50299182 Is our new policy to stick our heads in the sand and pretend that CVEs don't exist? Because I'm not getting behind that!
76 comments
[ 0.28 ms ] story [ 35.1 ms ] threadImagine abliterated frontier AI models reach >20,000 TPS. As I said before [0] you are going to have to hire specialized technical bodyguards from the senior to senior staff level using AI agents to protect your entire business.
Otherwise you are cooked.
[0] https://news.ycombinator.com/item?id=50310312
[0] https://www.youtube.com/watch?v=9OahTki4cPA
Is that really the best way to do things? I’m not really sure what an agent offers here does it just look at logs and do actions? I would not really feel safe if that’s my service’s answer to security.
I could see it maybe as a really smart filter to block ips that are sending probing requests but I don’t really see that as a good reason to have an agent be sitting 24/7 looking at logs.
Thanks god, you are not a wacko thinking qwen3.0-1.5B-abliterated would hack a goverment.
[0] https://status.linode.com/incidents/8dqqkh4d9y2c
[1] https://status.linode.com/incidents/zp71823t318w
[2] https://status.linode.com/incidents/wq8zlhk736fk
Ask HN: Who didn't get hacked today.
I never even heard of it.
https://docs.cloud-init.io/en/latest/reference/modules.html#...
And then set up the subcontainers with their own cloud-init:
https://docs.cloud-init.io/en/latest/tutorial/lxd.html
So I am guessing this might appeal more broadly.
I don't see how web services proceed like business is normal.
- Tell HNN: pets.com was *not* owned today
- Tell HNN: about:blank was safe, too
- Show HNN: my new gopher protocol project—no currently known CVE 10's
If you are running services and don't yet have an easy way to react to these things with swift process defined, I hope you don't like to sleep.
This isn’t a new thing for Linode. A customer service panel vulnerability in the past was the reason Bitcoinica got hacked in the early bitcoin days: https://spectrum.ieee.org/thousands-of-bitcoins-stolen-in-a-...
Did you know that GO defaults to listening on TCP connections with MPTCP, so anyone running GO programs listening on the internet (Caddy, traffic, etc) and are on an unpatched kernel now have a way to from attackers to get root on that box?
Even if you're not using MPTCP, this functions as a DOS attack.
It's a very bad CVE, so yes. Follow the advisor and patch.
And even patching the ‘small’ cves will take us for a spin as it already has. Then there can be specific low hanging fruits that chained together make a splash.
I think people would keep paying, even for pause.
Your intuition that this can be balanced out is wrong. When you have a CVE in your TCP stack, your vault unlocks and you're fucked, it's done. My dude, there is never going to be AIs scanning every incoming packet for signs of a 0day, and deciding whether to, uhh, cut it up with a lightsaber and shoot it with a laser pistol. Your options are to have a stack so secure that it can't be targeted, or to disconnect all network cables.
It is possible to write more secure software[0], companies just choose not to prioritize security.
[0] https://cr.yp.to/qmail/qmailsec-20071101.pdf
Don't use MPTCP is the first mitigation, but my understanding is this might be exploitable without MPTCP.
I would stick with a "too big to fail" institution instead of splitting across multiple.
If you want a real "backup" you could ask for physical stock certificates or get physical gold or something else that cannot be taken from you by computer hacking.
I think these banks will probably be able to hire equivalent people in the private sector though and might not need the government help.
Simply put I don't think dropping in smart people in dysfunctional organization is going to have instant results. It will take a pretty long period of time to make them aware of all the systems with holes.
In the meantime hackers would be running off with actual information and money. More than likely I see banks turning off all kinds of systems in a panic and people with their money there being in an even bigger panic.