But the Dutch government websites do generally (I hope) have users, which is one of the ways the "I need to hit the same php worker, uninterrupted, for about 10 to 30 seconds while absolutely no other traffic hits the site AND I can isolate the one worker from the pool" approach to exploiting fails.
Luck is not a security boundary, they should patch their servers, everyone should. It's not a good thing to have shitty memory management on your server.
However, is this really that exploitable in real world conditions? (no)
"There does not exists" and "I've not heard off" are different things. This is motivated reasoning on your part to tell the story that this isn't a big deal.
Most variants require a few requests to infer some settings, and then they can spam their arming-stage, and then their exploit stage. Granted, its not quite as easy as that, but hitting _a_ worker twice on a server once is not all that lucky.
So according to you they could have just imposed an upload cap instead and been safe? If all the cybersecurity experts had given useful advice like that earlier in the week people could have been safer and less panicked.
The only thing I read about why Dutch gov websites "went down" was as a precautionary measure against this exploit (awaiting the patch), not because they actually got hacked.
Can you link any reports about any site being afflicted by this exploit? (Dutch government website, or other)
The parent comment is implying PHP servers arent at risk, because default nginx limits requests to 1mb. I'm saying that is a bad argument, and a lot of real websites that went offline voluntarily, like the dutch gov, would have endpoints that accept >1mb requests.
Putting https://stealmybitcoin.com/ behind a default nginx proxy and giving it no features is a toy PR stunt hoping to salvage PHP's reputation.
Hi LiveOverflow, I'm Elias and you made a video reply to my blog post. Why did you delete my comment on your video? (EDIT: Youtube fucked with my comment, not LiveOverflow! Editing to put this at the top of the thread so that it is read)
Your second exploit that you slopped out - after someone pointed out your first exploit depended on reading the container's memory via `/proc/pid/..` - also depends on an extremely contrived thing outside of a lab.
To get around the fact that you need to hit the same php worker over and over again (this is another way the LLMs were stacking the deck in their exploits; so many of the exploits online had php running with one worker - which is an unbelievably unrealistic precondition. That, and with ASLR off, lol.), you hold WORKERS-1 connections open indefinitely (lol) and then need to sequentially, and stably, hit the same php worker for 10-30 seconds.
Of course, luck is not a security boundary, everyone should still patch their servers.
But how is it a realistic precondition to have a server in which you are the *only* person making requests for 30 straight seconds? Even a single googlebot pinging the website would fuck your heap grooming up, let alone regular users, or another person trying the exploit out! Similarly, Holding WORKERS-1 connections open indefinitely so that you can isolate one php worker is extremely unrealistic in practice.
Yes. this is a vulnerability, it is bad. No, it is not easy to replicate in real world conditions. The LLM gets you an "exploit" locally. How on earth did you not extrapolate from the LLM decieving you the first time, and think for some reason the second time the conditions were 100% truthful? Did you even read the stdout from the exploit you wrote?
I really like your work man, you're a great inspiration, but what the fuck are you doing? And why delete my comment pointing this out?
I am not deleting any comments ever. I just checked, YouTube held your comment as "likely spam". I accepted it now and pinned it, I always appreciate it when sb is reaching out with a reasonable comment!
It is an absolutely slopped exploit, hahaha, I fully agree. But that was exactly my point. Memory corruption exploits over the internet have been historically extremely difficult. And it took many days of building a PoC. And this is clearly different now. And of course there are a lot of practical challenges, but that does not make it unrealistic vulnerability to exploit.
> But how is it a realistic precondition to have a server in which you are the only person making requests for 30 straight seconds?
I think it's VERY realistic. My random blog from 2023 https://raisistance.com has hours or days without requests. I think a majority of (old) PHP sites barely anybody visits. And there are tons of random company or government website without any requests at night or so. Also depending on the setup, there ar methods to deal with concurrent traffic. For example Apache (see here: https://x.com/LiveOverflow/status/2113555517339480284).
I think it's clear so far that it's not mass-exploitation material. But once you are more targeted and have a random LAMP stack, it becomes very realistic.
Would you setup a random vulnerable php version from Docker image like `php:8.5.11-apache` on a server, put a flag there, and give me the IP? For me this is a realistic setup and it's fairly easy to exploit.
Hey, good to hear that it was just YouTube holding the comment. I really thought you deleted it which was audacious. I'm now much less mad :P.
I think we can set up something interesting here, I'd be down to set up a little "PHP2Shell Olympics" as we change preconditions and see how well the exploits fare. I am pretty sure our opinions of "this can be exploited, and LLMs can aid a rapid exploit of previously-complex ASLR nonsense" and "it's difficult to do memory corruption over the internet; php2shell is not a log4shell moment" are mutually compatible.
My contact details are on the bottom of the blog post.
Really, your request log is empty for hours at a time? I find that hard to believe. Even old blog sites still get an insane amount of bot traffic. I have sites that are only barely in googles index and it’s pretty rare to see anything less than 1 req/s (99% bot traffic)
Yes! Even my own stuff sometimes goes a few hours to a day without a single request, the internet is either very slow and dead or super active, no in-between, I think it heavily depends on who indexed you and how big your site looks like or maybe identified stack (for vuln crawlers, i'd wager if they see a wordpress marker they'd put you on some list to get retried by the swarm) or something.
My datamining project for a game for example, that noone besides me and 2 friends use, gets thousands of requests per second, because all the AI bots found it and try to index every single page (thousands of ssr pages) and they repeat the requests when my cache-control headers expire (ie: when i consider the old page stale), so it's an endless stream (I dont mind the challenge of serving all of these at high performance xD).
My main website, in the same domain (not a subdomain), barely gets any requests, it only has the one main index page with a single link to the datamining subdomain, that's about it.
From my raisistance.com blog 10 October 2026 00:05:56–25 October 2026 09:25:02 UTC:
- 95.3% of seconds had zero logged requests
- 11,582 of 22,159 complete minutes (52.3%) had no requests. The longest quiet stretch was 27 minutes 26 seconds.
- 3,283 continuous quiet stretches lasted at least 120 seconds. Together they totalled 225 hours 27 minutes 27 seconds—61.0% of the logged timespan.
- Excluding Apache’s internal dummy connections: 3,305 qualifying stretches totalled 227 hours 1 minute 51 seconds—61.5% of the timespan.
- Of 38,389 seconds with exactly one request, 14,692 contained an Apache ::1 internal dummy connection. Others included homepage visits, WordPress login requests, crawlers, and scanner probes.
All the "working" php2shell exploits on GitHub and X are like that. There are ZERO publicly available RCE exploits for php2shell that actually work remotely on production deployments.
All of these RCE PoCs are local and rely on unrealistic setups because LLMs couldn't achieve RCE without resorting to reward hacking.
I find it offensive that your blog is named after thelastpsychiatrist when the writing is not even close in quality. Come on man, I expected you to start swinging. Are you building an agent tunnel? Get outta here.
Expands into a document for _who_ to read? another LLM to re-compress?
It's baffling you people are in control of such a strong product when you are obsessed with this intellectual pornography; wow - look at how smart it made my thoughts look (n.b. look, not read). Don't look too close. And certainly don't ask me what it means.
Guys. Guys, you got it all wrong. We don't want to ban open-weight models!
We just want to ban the competition guys! Very different.
--
The ridiculous anthropic/openai strategy of selling shovels at a loss in a gold rush isn't going to play out, and the hilarious thing is that these AI companies are going to create tons of value and _capture none of it_.
Their only path to profitability is if they get to capture it and they're going to do everything to do so. Put it this way: *all the blog posts that Anthropic and OpenAI are putting out are DESIGNED to scare you so that you let them capture the market*.
...and "distillation attacks" (hilarious framing of "saving the output of our models")... Whatever.