44 comments

[ 0.26 ms ] story [ 33.1 ms ] thread
This one is both awesome and scary. We are living in a black mirror episode, where one well-crafted LLM prompt can get you $500k or the ability to hack into 500M websites :)
Nice balance between practicality and ambition. Curious how it holds up with real-world scale.
The surprising (and possibly untrue) thing is the high price of canned vulnerabilities. WordPress is known as the remote root shell with a blogging feature.
Presumably they don't pay $500k anymore...
So they spent the $25. But the real question here is did they get the $500K?
Can RCEs like this still be sold to exploit brokers like Zerodium? If so, how? Asking for a friend
Just like that, the 0-day black market is experiencing a mini black Monday.
Interesting write-up and I do think LLM assisted/powered exploit disclosure is a real concern (I've been able to get models to create container breakouts from Linux LPEs relatively quickly).

One thing I'm surprised about is that GPT-5.6 didn't block that prompt due to guardrails. My experience is that GPT-5.5 and up does not like offensive security work (similar to Opus 4.7+/Fable).

I didn't notice it but I'd assume that the authors have some level of cyber approvals from OpenAI to relax the guardrails a bit.

> Is GPT5.6 Sol Superhuman?

This is not a simple y/n question. Computers have been superhuman at playing chess for decades now. Reading this article, I guess they are superhuman at understanding code now as well.

I like the idea of not crediting the person who posted the bug but to the LLM that found it. People who find exploits using LLMs should never get a reward or credit.
This assumes those who'd pay $500k don't have the skill to use GTP5.6 for the same purpose themselves?
What was the harness used? And yeah surprised about such prompts not being downright blocked, even with the cybersafety verification"
The cost of business nowadays? At what point does it become apparent that in today's world software needs continuous pen testing and scanning? If you don't your attackers will.
The author lost me at the last bit where they started using weird names for the posts. Why would you make one ID O and the other ID 0? Why single letters and not EMBED_01? Why seemingly random letters instead of ABCDEF? Does OCPDST stand for something?
On another note - who needs WordPress in the age of Astro and LLMs?

I mean, it's not a taunt, but a serious question - do people keep WordPress because it used to be the easiest solution to set up years ago, or are there still clear use cases where one can argue it's the best solution?

That was an incredible writeup! Chapeau to the author - thanks for taking the time to do a writeup.

When Anthropic claimed Mythos chained 4 or 5 bugs to achieve sandbox escape and found bugs in core software, it sounded like bs. But here we are 2 months later seeing what they meant.

Cybersecurity was always a hard sell; security flaws were invisible - by contrast a fence with a hole is visible to everyone - anyone can ignore the locked gate and walk through the fence hole. With cybersecurity a hole in the fence may go unnoticed for years, maybe forever.

LLMs level the field. We will all benefit from more secure systems, a few people will get a lot of egg on their faces, and it will end the malpractice of underinvesting in software security to get a product out the door.

wordpress is so shitty - though it runs the majority of the web.

people think PHP is shitty cz of Wordpress.

at a certain point in time - people need to move to better ecosystems painful as that may be.

Historically, a lot vulnerabilities in PHP projects were because of PHP itself. Things like register_globals, remote includes/reads using functions that supposed to read local data and so on.
There is no evidence that $500k has been paid or would be paid for an exploit like this one.

Given that the article says that prompts are modified like they are holy scripture, perhaps sell the prompt for $500k.

The author works for https://www.assetnote.io/ , which has AI products for automated scanning.

What do you think the venn diagram looks like for people willing and able to find things like that prior to LLMs and also sell them to a broker, and are also stupid enough to flaunt a massive flashing "arrest me!!!" sign

Closest you're going to get is something like those kids in florida who just got wrapped for putting malware into steam games and draining peoples accounts. They were going to get caught anyways but it would have taken a lot longer to build a case against them if they weren't flaunting it on socials

(comment deleted)
Ok, we've taken $500k out of the title above.
I feel like I've seen plenty of non-AI, pre-2020 SAST tools catch SQLis like the one mentioned here, and if nothing else, then a code review ought to catch it. Is WordPress not using code reviews and/or SAST?
They would not catch it, it requires combining multiple vulns.
I am so done with FOMO writing. Sure man, you found one with $25. With $25 plus your entire industry domain specific knowledge of where to look, of how to probe, of what else you may have accumulated and collected over the years of working within this industry. Let's stop with the gambling narrative and the illusion that we are all missing out.
Should also note the math done on the token costs. $25 of subsidized tokens because he's on a subscription plan.
Does Sol allow this kind of research by default or is this a "look at me I'm on the cyber research allowlist" post?
(comment deleted)