I took a look at the Ghidra ones (because I use Ghidra), and I'm unimpressed: https://github.com/bikini/exploitarium/blob/main/ghidra-12.1...

The first requires being able to overwrite binaries in the Swift tool directory. Yes, if you overwrite binaries executed by ghidra, you can trigger code execution. This is not a surprise.

The second, idk, I'm not familiar with TraceRMI (but it's probably worth noting that "RMI" stands for Remote Method Invocation).

The third is not a vulnerability in the slightest, they just demonstrate that native 7zip parsing code is reachable. Maybe there is a bug in the 7zip parser, but without that it's meaningless.

> ...Ghidra

Ah, MS07-052[1] ("code execution leads to code execution") strikes again. These sorts of "if you allow arbitrary code to be executed, code will be executed" "vulnerability" reports seem evergreen

[1]: https://devblogs.microsoft.com/oldnewthing/20070807-00/?p=25...

A glance at the nmap one seems potentially high severity. It might be a nothing in practice, but it being around parser code means the chances of preparing something to jump around are pretty high.

There'd be a certain irony being able to reverse shell anyone doing an nmap scan. If i had infinite tokens i'd throw claude on writing an exploit and dig through the history who made it possible because - if we take a moment to wildly speculate and assume it can ACE - this is the kind of bug an intelligence agency would love to have: Add a few ipv6 packets that then edit the trace being observed if the observer uses nmap / get access to any researcher pc who uses nmap.

These kind of tools have always had a broad attack surface. I've assumed state level actors already have exploits for them, mostly based on when I've used such tools for mundane network maintenance tasks and somehow do something that triggered an old-fashioned segfault.

Wireshark dissectors (protocol decoders) are basically all written in C, and anyone sending packets can pick a dissector.

I'm more worried about AV software. Code that also needs to be able to parse a large number of file formats, opens every file that enters your computer through one of many pathways, and generally runs at a high privilege level. A huge attack surface that's easy to reach and with far reaching consequences if it can be exploited. Add to this that it's in wide use, often even mandated by corporate IT and its recipe for disaster.

In theory the parsing could run at a low privilege level subprocess. Root/admin is only needed to get the bytes.

> I'm more worried about AV software

Media codecs pretty much, single-handedly even, drove about a new era of defenses and mitigations in Android: https://blog.isosceles.com/the-legacy-of-stagefright / https://archive.vn/x3d0Y

>There'd be a certain irony being able to reverse shell anyone doing an nmap scan.

Every TV / movie hacker has known about this.

Was just thinking it would be hilarious if these were all known CVEs hiding the next Shai-Hulud inside of them and waiting to compromise security hobbyists rushing to download them.

It wouldn't be the first time!

The design purpose of Windows Sandbox :)

TIL about Windows Sandbox. Not a Windows user myself (it's the Year of Linux on the Desktop™!) but this sounds immensely useful for technical and nontechnical users -- sort of a super-disposable container with UI? https://learn.microsoft.com/en-us/windows/security/applicati...

The short lifetime of it really seems to limit the usefulness. I had a look at it before and really struggled to see what I would use it for.

Thank you. TIL. Shame you can’t run multiple simultaneously, but still it looks great!

So I've been wondering about whether there's a new vector that can come out of this? Or perhaps new vector isn't the right way of putting it, not a security expert :)

By mass sharing these kinds of gaps and utilising mythos tier LLM's ability to find and combine multiple disparate bits of information together, are we increasing it's capabilities and versatility?

Ghidra one is pretty weak, but I checked out the ones that were interesting to me (c-ares, libssh2, ffmpeg) and they seem to all work as of the latest upstream commit. Weird

> Yes, if you overwrite binaries executed by ghidra, you can trigger code execution.

> but it's probably worth noting that "RMI" stands for Remote Method Invocation

This reminds me of someone submitting a (clearly vibecoded) vulnerability report claiming to have found a way to execute arbitrary SQL. The project in question? An SQL server... https://github.com/tursodatabase/turso/pull/4322

lol, that's great...the "vulnerability" isn't even in turso itself, it's a helper method inside a unit test.

actually that is a valid vulnerability if it wasn't in test code but the correct fix would be to enclose the table name in "" with escaping

The Turso example is a little ironic because their homepage brands them as a product intended for use primarily by AI agents.

The Gitea one looks marginally interesting, but is probably not exploitable in practice (unless Gitea or whoever else isn’t properly isolating jobs on dedicated VMs). I suspect GitHub Actions has similar behavior and is not considered exploitable because the user is assumed to already have local, non-namespaced root access.

Gitea action runner has a bunch of different ways to setup and doing the isolation properly looks tricky. The documentation doesn't provide any isolation tests to administrators, either.

The biggest mitigation is that gitea documentation discourages you from using action runners from untrusted users. Not flawless security, but it's something...

> The biggest mitigation is that gitea documentation discourages you from using action runners from untrusted users.

This recommendation seems incompatible with third-party collaboration, at least on its face!

The idea is you first review PRs from external contributors before allowing the CI to run on them.

Potentially, but for many projects things like that are tools that you want to control access to anyway. Anyone wanting to update the CI/CD process who isn't a trusted part of the project should be having their changes properly reviewed by someone who is anyway, at which point the reviewer is the trusted user not the random external entity.

I don’t disagree with that, but I think GitHub has shown that projects want to have their cake and eat it too. GitHub has also shown that it’s incredibly easy to design an insecure CI/CD that satisfies that goal, but I see that more as a symptom of them being first-to-market rather than an inherent quality of the problem.

Wait, isn't this about protecting the machine running the actions? If someone hosts a project on Github and allows anyone to run actions, it's Github's problem if there's a vulnerability to exploit. It's their installations that are going to get compromised, not necessarily the project's data.

I'm no expert on any of these programs, but that's kinda the problem, isn't it? No single person is an expert on every codebase supposedly exploited in this repo.

After a bit of research, the Firefox one seems plausible to me. But, I haven't actually tried the POC. The explanation about the private-data and untrusted-input flags is plausible but I'm not an expert on Firefox's internals, maybe that's not actually how it works.

This just sucks, all around. Are we going to need every open source project gawking at the same repo full of stuff that has nothing to do with them, on the off chance that someone discloses a vuln that does have to do with them? Is this some kind of performative complaint about high friction in responsible disclosure? Well great job dickhead, you've just made a system that's even worse. Nobody benefits from this. Yuck yuck yuck.

I actually prefer them being public than in some governments or corporations toolbox

> Nobody benefits from this

Disclosures always enable more secure software to theoretically exist,

even if nobody follows through creating it.

They often do.

The inputs are truncated to prevent malicious prompt injections. The PoC ignores this by using a fake model which is easily convinced.

>The first requires being able to overwrite binaries in the Swift tool directory.

Does it? Or does it need to be in the same directory you invoked ghidra?

I immediately saw the Ghidra one and was thinking: huh?

The point is that anyone looking for zero days has them in spades, in this age of LLM use.

So, knowing that bad actors have an unending river of cheaply acquired zero days, the best response is to publish them so that maintainers also have access to them. Existing methods of slow disclosure cannot keep up with the AI firehose.

It’s ugly, but it will force needed change. A thorough AI red team effort is the lowest bar of releasing software responsibly in this day and age.

If only the AI tools didn't shut you down every time you were trying to red team your own tools. I've had to come up with all kinds of workaround scenarios, effectively bypassing the AI security processes in order to stress test my own systems.

You don't need to be pissed off to decide that immediate public disclosure is the best option.

Ok, I don't know their emotional state. Fair point.

Maybe I'm projecting my own biases ;-)

Meanwhile, some dude was just playing with claude and accidentally made his repo public.

Approaching the maintainers would be ideal but time-consuming. Disclosing it like this is neutral I guess. Better than selling it in the darknet.

I do wonder though: if you can tell the AI to search for vulns, can't you also tell it to contact the right maintainer for each one found?

Unfortunately, yes.

7zip has been known to be buggy ,its very likely.

code execution js code execution, if reached through some bug or executing code that was not intended to be executed its bad, even if the mechanism is kind of obvious and trivial, it still can lead to unexpected code to be launched.

the repo also notes low quality of some POC like ones you noted.

its correct to be a bit wary and i wouldnt call it some crazy 0day dropping account or anything, but bugs/vulns are bugs an vulns and simply because you are not impressed by their complexity, it does not reduce them entirely. just makes em lil less scary.

the ghidra one, it reminds of things like unquoted service paths in windows services. its a silly thing and clearly its bad but it still happens and gets companies pwned :'). a lot of companies use ghidra actively and if you imagine what kind, you'd hope they will not allow the overwriting of those swift binaries ;p. some pentester bound to have a laugh.

Went over a few of these with a pretty keen eye, and they aren't that particularly interesting. The Docker one is just a weird bug, it's not a vulnerability, and certainly not a "0-day" (which is a pretty loaded term and people expect bad stuff to happen).

The nghttp2 nghttpx one is more interesting, and could potentially be used for phishing, but it's very hard to line up properly because the request queue is non-deterministic so basically impossible to target a specific victim (assuming proxy traffic).

The VLC one is just a straight-up crash/bug. And VLC crashes all the time when using weird codecs, so that's nothing new.

Am I missing something here?

Do you feel the same way about seeing "Segmentation fault (core dumped)" in the terminal from some other arbitrary program?

If not, why not?

Because people often use VLC to view untrusted content.

Video is a great vector for distributing malware, especially sought-after grey area content like porn, conflict videos, celebrity leaks, pirated films, etc. Not enough people pay attention to the impact of video as a vector for compromise. All downloaded video should be sandboxed!

Right, this is why video parse / decode ought to be sandboxed. Writing secure code for these formats, especially in C, is really hard. I just sort of glanced at the bug in the repo, but it sounds plausible. It certainly wouldn’t be the first of its kind.

And it's my impression the code is often in assembly, which is even less structured.

> I mean, that's how people get hacked.

...when was the last documented case of an in-the-wild hack targeting VNC?

VLC != VNC

Oops! I was on my phone. I meant VLC.

I recently used a pretty well-tuned LLM to find ~500 safety bugs across the Rust ecosystem. Most of them are minor, and even major safety issues in Rust usually mean "it's possible to accidentally use this API in a way that is broken" not "this is directly exploitable", but I didn't want to just file LLM output as issues on these repos.

I very briefly considered doing something like this: if I just post the results on the internet, people can crowdsource filing issues and working on fixes. It's certainly not the nicest way of doing this, but on balance I'd like these issues to be fixed eventually.

I ended up not doing that and am instead filing a couple issues a day because it's not that much of a burden. This was an experiment that was much more successful than I expected, so I didn't budget to spend this time, but it's also not a huge deal to slowly do it.

Can you at least provide some links to examples of the issues you’re filing?

I’ve seen so many claims of people who used LLMs to generate hundreds of issues that turned out to be full of hallucinations or non-bugs being described as bugs that any claim like this needs some real evidence. Saying you found 500 safety bugs in Rust but that you’re casually only filing a couple of them makes this hard to believe.

The readme was recently updated.

> In regard to AI usage, my fuzzing workflow was automated by AI with a strict harness. I used GPT-5.5-3-Codex-Spark for ALL the fuzzing, as barely any "thought" is necessary when provided with an efficient harness. Contrary to the growing narrative that I'm just some random child burning tokens, I DO actually have a degree in the subject and have published multiple papers on fuzzing methodology. I spent years researching and developing new tools and ideas for how to fuzz. You do NOT need a SOTA model to help you identify these issues, I promise!

Any thoughts on what type of harness this would be?

0-days-vibes-vulns? There should be a new category, for spotting and handling the em-dashes of this brave new world of vulns and making the old fossils like me only picking my head up for the old painfully still hand-crafted artisanal ones instead. A kind of label, like free-range for eggs, in sum.

Yes, big pet peeve of the new world. Every em dash is apparently an AI trigger. Back in my day, they were a sign of great respect within my people.

I used to be an em-dash user, but now my opinion is that I’d rather be perceived as someone who does not want to be confused with an LLM. So I’ve changed my writing style.

My feeling is that my writing doesn't sound anything like an LLM, so if someone thinks I'm an LLM because I used an em-dash, that's on them. That, or I royally screwed up and need to do a better job as a writer. At least with today's LLMs.

> if someone thinks I'm an LLM because I used an em-dash, that's on them.

I'll go a step further: I think I'd rather actively filter out people whose AI detection is that naïve.

hmm maybe if we wait longer, LLMs will learnt to be like "none-em-dashers", and we "em-dashers" will prevail?

There's also the "paradigm shift" and "a fundamental Z of X". Those really ding my slop'o'meter.

I don’t give a flying fuck what people think. Most colleges copied or adopted my (for a few semesters) school’s style guide, so LLMs are essentially copying me, and I won’t change my punctuation usage because they suck.

Yeah, I get it, they do suck. It all sucks.

But people at work who are copying responses from LLMs into emails to others also suck, and I want to distance myself from them as much as possible. I'm kinda hoping we will eventually have a wave of "what the fuck are we paying you for if you're just copying stuff from an LLM to Slack" firings.

This is a pointless and infinitely losing arms race. LLMs will learn to use hyphens instead of em dashes, and so what? You’re going to start using em dashes again?

Just focus on not producing slop.

At this point, I believe the LLM "style" is on purpose. Perhaps the labs want to be able to distinguish their slop from human content for future training; I don't know. But it feels very deliberate that they've kept the style so consistent for so long.

I agree. They're voluntarily adding fingerprints to images so I expect the default voice is intentional and it wouldn't surprise me at all (though I have no evidence of this) if the output text has a fingerprint stenographically embedded in it.

In a video of Hank Green, he interviewed an AI expert (if there even is such a thing), and he said that in the thinking part of the conversation, LLMs seems to use code language to communicate with itself, like in the usage and ordering of words, and such.

I think that there could be even more then a fingerprint in those messages.

It’s this video: https://youtu.be/5CKuiuc5cJM?is=9VQ1FCxY_X3eNm-b

Warning: They anthropomorphize a lot in this video, but I get it… the words exist, why not use them.

I'll take this opportunity to repeat that the natural language interpretation of thinking traces don't appear to be "real" by any reasonable definition. Even if they can at times be useful (at least seemingly). There's research demonstrating the usage of arbitrary symbols, even just repeating a single symbol, leading to a similar improvement in ability. This makes sense if you consider how the attention mechanism and KV cache work as the sequence iteratively grows.

Basically we optimize the models to produce output with certain characteristics but that doesn't mean that what we see is the whole truth or even that the relationships in the underlying system are structured in the way that we might expect.

Sounds like a case of the Mondays, Mike Bolton

YYyeah… that’s definitely the correct reference, Bob.

Ok— I don’t give a flying fuck if people jump to the conclusion that I am an LLM. And that concludes my willingness to engage in those pedantic semantic antics.

They all say that, until someone that matters mistakes their writing for LLM slop and they lose something of value due to it.

The purpose of good writing style is to communicate more effectively.

Sometimes you’re trying to communicate education or class. Often that’s not the main goal.

Shakespeare was written for the masses. Hence being full of dick jokes.

Similar story for Chaucer, and so many others. I don't think good writing, things we appreciate so much it lasts generations, has much to do with signalling education or class.

> The purpose of good writing style has always been to signal education and class

Sure, that’s why there have never been any authors that became famous despite being poor and deliberately writing with that affect.

Good writing style does connote good education, and in environments where being upper-class bolsters social standing, some people flaunt it to signify class, as they would with any other wealth signifier, like expansive shoes.

I am a union tradesman— the third generation to work in manufacturing in this area. Affecting an upper-class identity diminishes social standing in my environment. Having a lot of money, definitely doesn’t. My dirty work boots probably cost as much as many of the trendiest shoes on the market, and the guys at work know that and admire them… but my wearing them doesn’t signify class. Similarly, you can use good writing style in a way that shows you went to a good school and paid attention without wearing it like a Harvard Business School fleece.

It’s fine to use em-dashes — just be srre to add typos.

You can also have the em-dash itself be a typo, e.g. using the figure dash ‒ (U+2012) instead.

I just use a hyphen. It's easier to type, easily understood and you won't be accused of using an LLM.

They're just so handy! I do think LLMs tend to use them in a specific way, though.

So maybe tweaking your usage (ex. no spaces around them) or using a technically incorrect en-dash might offer the desired effect while subtly signaling that your message isn't AI-generated.

I still use them — mostly for pauses — but I'd like to think my voice sounds distinct enough from an AI that people can tell.

I've only ever been using "regular" dash, a minus, for that. How do you even type yours? If I ever needed differently-sized dashes (and I don't know the difference between them) I always used wiki to copy them.

(disclaimer: I feel like this obsession with dashes is special to native English speakers, which I'm obviously not)

silly specific: the minus sign is a separate character. The dash equivalent is the en dash (–), versus the larger em (—) and smaller hyphen (-).

The en dash is also used in things like scores (3–2 Turkey), votes (the bill passed 58–42), or connecting words where the second part is longer than one word (the Australia–New Zealand alliance.) You can remember the latter as, "a hyphen isn't big and strong enough to hold on to more than one word.

If you're on a mac, pressing Option+- is the en dash and Option+Shift+- is the em dash.

My first language is Dutch (Flemish). I didn’t even know there were three different dashes. The em dash is something I didn’t know exist, or were used, until llms came about. The hyphen is something we use when we make a list, that’s just the minus symbol right?

So em dashes are for pauses or highlighting things I guess? The en dash you explained in your reaction. Is there any other use for the hyphen except for making lists?

There aren't three different dashes. Unicode 15.0, table 6-3 (Unicode Dash Characters) lists 25 different dashes. https://www.unicode.org/versions/Unicode15.0.0/ch06.pdf#G612...

Only a small subset of native english speakers. Most don't use dashes at all, of those that do most just use minus for everything, some exceedingly small group cares about typographical details and thus distinguishes the different sorts of dashes.

It's an attention to detail thing that you'd definitely want to get right in a physical textbook or the like.

That's the same in my native language too I guess. It's used in books, sometimes in media, but not really in any casual setting I would say.

Depends on your OS. Mac is the easiest, it's just ---, Linux depends on your distro, if it uses KDE, it's <right-win>--- —. Windows is a little awkward, I think you need <right win>+the code point.

I have mac, typed this --- again --- and nothing? Layout says U.S.

edit: another comment gave a mac shortcut – — - <--- one of these might be it

Option-Shift-Hyphen.

Just --- only works when you have the text replacement thing on (the same thing that turns (C) into ©).

It’s an obsession with literature and/or typography nerds specifically.

Option-shift-hyphen types an em-dash, option-hyphen an en-dash. You can also hold the hyphen key (on a Mac or iPhone) and it will allow you to select either. Em dashes are used—like—this—as something spiritually akin to a parenthetical. En-dashes are used within ranges: Feb 14–17.

I for one am striving for clarity and couldn't care less about being confused with AI.

However I've only ever used regular dashes. How do you type an em-dash? Is it OS specific? I've taken to using Emacs insert-char with a list of frequently used ones in my scratch buffer. My memory for Unicode is unreliable.

Keyboard layout specific. Macs with their default English layout use “option-shift-dash” which is really easy to remember (and relatively discoverable, as such things go) which is why using proper m-dashes (not just double-dashes) used to be a strong indicator a poster was using a Mac, before LLMs took the character over.

On iOS you type it by pressing dash and holding until alternative options come up, same way you type e.g. accented characters.

Macs have two possible ways. If you have key repeat enabled, option+shift+dash. Some newer Mac users may have the mode on where holding a key pops up an iOS-style bubble of alternate options, in which case they will just hold hyphen.

That "new" feature has been around since at least 2011.

You can also just type two "-" minuses on iOS. So "--" will auto-convert to "—".

In emacs, c-x 8 RET prompts you for unicode character names (or hex) so for rare use you can just spell it out. There's also C-x 8 _ m for em dash and C-x 8 _ n for en dash. (Hit c-x 8 c-h to get a full list of those bindings, like any normal secondary map - they're about as idiosyncratic as the XCompose bindings, but you might find some of them "stick" in your head better (I personally like "C-x 8 1 / 2" better than "Compose 1 2" even if it's a lot more typing...)

Ooh, thanks! I blindly type c-x 8 RET, then finding myself stumbling through 9 different entries for Phi, but I hadn't explored the prefixes. I just realized I could create a custom which-key entry-point with my favorites in a prefix.

Also just learned about compose key apparently, and I noticed that I can program this split keyboard I'm using to turn that into a chord, anywhere!

Then an LLM told me that I can 3D print my own custom keyboard with 32 programmable layers. Everything is an infinite rabbit hole these days, how wild.

>In emacs, c-x 8 RET prompts you for unicode character names (or hex)[...]

Which is super useful for hard space - non line-breaking space - so that one letter words don't appear at the end of lines.

Huh, I always wondered what that was for...

Macs have a native way to do dashes: option- hyphen for en-dash and option shift hyphen for em-dash. On Windows there are some application-specific ways that make sense, e.g. in Office, but outside that you’re on your own and have to use the “hold alt and type the character codes” method! Or charmap.

> How do you type an em-dash? Is it OS specific?

On Linux X11 at least, you can enable the Compose key and then press `<Compose>---` which results in — and `<Compose>--.` which gives you –

“⎄---”

I now use "ASCII em-dashes" by using two hyphens -- like this. Or--if you prefer no spaces--like this.

Nah, I’ve started noticing people doing this replacement automatically in LLM output. I just try not to write with dashes anymore.

the nn dash remains the goat. the arg dash

Don’t you love when your arg dashes get autocorrected to emdashes? And by love I mean hate with the fiery passion of a thousand suns.

Agreed. On an iPhone that’s the easiest way to type an em dash and consequently the easiest way to fuck up trying to write out a command line example.

I like the cut of your jib.

same, or I use a semicolon

semi-colon will have me thinking you've typed up your comment in Microsoft Word

I use a semistoma due to medical reasons.

Code switching in the post LLM era.

I'm a secret invisible &shy; soft hypenator. I like to break words in whacky places that change their meaning only when they need to be broken. Like Democ&shy;rats.

https://en.wikipedia.org/wiki/Soft_hyphen

It's a perverted expression of hidden passive aggression.

It looks aesthetically nicer. It was also a bit of a signal that someone took pride in their work and so helped that way. It's a bit like whether your tradesperson cleans up after themselves. Technically sweeping up the dust after installing a kitchen cabinet doesn't actually mean anything for the quality of the kitchen cabinet installation, but in practice putting the effort into the presentation correlates with putting the effort into the actual work.

I understand the sentiment with the tradesman example. I would argue that there are a multitude of ways to care about or take pride in one's trade and emdashing the floor is just one of them.

I think people who care about correctness and also read a lot automatically see the difference and it seems (and is) technically incorrect to use a hyphen where an em dash belongs. That’s really it. Kind of like you wouldn’t just leave out the apostrophes in your writing even though in most contexts they are not strictly necessary for comprehension.

Leaving out apostrophes isn't the same though, is it? If you leave out apostrophes you have syntactic effects while a (sortof)dash is still sort of a dash?

Its not exactly the same, of course, but its pretty analogous. It’s a minor, technically incorrect change that doesnt change meaning or understandability most of the time. Ive left out a bunch of apostrophes in this paragraph and Im pretty sure nobody will have a hard time understanding any of it. Yet it would bug me to actually write like this.

I don't know how this works nowadays but when I got taught to write (the pencil and paper kind) and specifically when I got taught syntax, it was "if you want to introduce a clause that explains or expands upon something that precedes it, you draw a small horizontal line" no mention of special chars, Unicode or whatsoever. So why is it so special other than being authors fancy?

Because ASCII minus instead of dash looks ugly. It's like using zero instead of "o".

An emdash means something different than shorter dashes

https://www.merriam-webster.com/grammar/em-dash-en-dash-how-...

i mean why use punctuation or any capitals at all, why not just blast words out in a stream of consciousness so readers know how yr thinking why even bother with speeling things write

Just because you don't care to use the proper dash doesn't mean everyone else doesn't. People have different levels of caring about different details. For the sticklers, there's even a special code point for ellipsis, … rather than .... (Four being correct, as one is to end the sentence.) Personally I'll just skip — entirely unless I'm in a trolling mood, though “sometimes” the right quotes are worth using. Special characters are easy to type on a phone soft keyboard, taking a long press on the relevant key, or if you're using any other advanced input system, so they shouldn't really be considered to be the mark of LLM input.

The real trouble is that people doing engage with the substance of the post anymore, and just shallowly dismiss a post as being vibe written, as if that makes any points raised invalid. Anti-intellectualism's always been cool among a certain crowd. Shame to see it spread but ah well, the propaganda's working.

I propose that humans use Unicode U+2E3B three em dash ⸻ it is an impressively long character.

> U+2E3B three em dash

I had to look up why this exists, and apparently it was added in Unicode 6.1 (2012) because some style guide required it, and using consecutive U+2014 em dashes isn't sufficient because that might not render as one continuous line.

https://www.unicode.org/L2/L2010/10037r-longdashes.pdf

let’s market it as “human dash”

And if it ever catches on with LLMs ⸻⸻ we just make it longer

Just write —(human)— to denote that a human wrote the dash. Just be sure to instruct your LLM to write —(LLM)— so readers know the difference.

Lets just use random number of dashes.

I'm still waiting for the interrabang to become compromised as well. They've already taken my em dash, and my use of the Oxford command, apparently, is an insant flag.

The thing is, in lots of contexts you cant easily type em dashes.

Google docs will convert “—-“ to an emdash but simple text inputs wont.

So when you see one in that context you have to consider the explanation. They copy pasted an em dash specifically, they drafted in Google docs, they know the unicode keyboard shortcut, etc. None of these are safe assumptions. And if it’s markdown you know it wasnt drafted in Google docs.

On Windows and Mac you can input an em dash anywhere with shift+windows+hyphen / shift+option+hyphen, respectively

in macos it is trivial to input em dash in any context

I still use them frequently. On iOS you just tap the hyphen twice, and it inserts an em dash—sorta like that.

I might like to see a collection of pre-2022 em-dash usage—a subset I suppose of the Low Background Steel category (https://lowbackgroundsteel.ai).

How many college kids are going to be flagged for using an em dash?

You completely misunderstanding the comment feels like an AI trigger

It’s so they don’t train on AI data, right?

The question is whether the m-dashes are surrounded by spaces or not. The spaces are utterly maddening. But yeah, RIP the mdash, who would have thought.

Those aren't even em-dashes and yet there's a huge thread talking about them.

It's interesting that Claude also over-uses en-dashes. It's very willing to create compound-noun-phrases, especially in that compressed-summary-paragraph it often writes. The 0-days-vibes-vulns that started this thread looks a lot like that, but it could be Claude directly, or just Claude's style influencing people who spend too much time with it.

Those are just hyphens, actually. En dashes are for ranges (e.g., 1–4, which is admittedly hard to distinguish from 1-4), not compound words. Point stands though—LLMs do love compound words and dashes in general.

..."and for the love of God, don't use M dashes when you write it"...input goes on for an hour droning about slop...

Are they all actually 0-day? I think a lot of them are from disclosed CVEs/code that were already fixed upstream. It often seems like the term "0-day" has lost most of its meaning today and people often use it to refer to any exploits.

Repo claims

> A single archive of public exploit PoCs and vulnerability research writeups. At the time I post these, none have been reported. Feel free to report them yourself and take credit for the CVE if handed out lulz. Please do not abuse these. I do this so to allure people into the field, and I've always found this is the most efficient way.

Which is roughly the definition of zero day. Whether the contents of the repo reflect the above claim is something else entirely.

> Please do not abuse these.

Reminds me of Jamie Wolf's joke about bestiality laws. Who are those for? What stops most people from bestiality is… not wanting to have sex with animals! For people who do want to, what, they won't because of… the law??

Who will this comment stop??

Well, it's a joke because the problem becomes apparent after you think a bit about it. The exact same reasonig can be applied to anything illegal, criminals are criminals because they don't respect the law, so you could try to say that laws are useless. Reality is, if something is illegal not only someone can be punished after the fact, but in some cases also preventive measures can be taken.

Regarding the comment, it isn't going to stop anyone. Most people will not do cybercrime because they're honest. Of the remaining, the risk of being sentenced to jail time will instead stop some people, even if not all of them.

Technically there are distinct crimes where we know that shifting penalties changes what happens but the impact tends to be on organised crime.

I mean I do actually think laws are basically useless. Good people don’t need em, bad people don’t listen to em.

I guess “bad” is excessive. I regularly observe traffic laws with less rigor the your average police officer would prefer.

As you're basically saying, the world isn't black or white. There's also a category of people that needs an incentive to behave. Also, without any laws, it would be problematic to punish "the bad guys". At that point, what you can or cannot do would be only a matter of relative power: who is stronger can do whatever they want to weaker people.

This is how it's always been. The law just produces another axis of strength that is easier to gatekeep.

Nope, it reduces the issue even if it isn't solving it entirely. Without laws and law enforcement, anyone with bigger muscles than you could break your nose just because they feel you looked at them the wrong way.

Physical violence is just about the last area we "need" laws because people really hate violence and will go well out of their way to avoid even the chance being subject to it. No matter how big you are you can't go around breaking noses without a good reason we all agree on because if you do that for very long a bunch of other people will gang up and break yours or worse.

To a first order, laws basically just codify how the government (the overwhelmingly dominant applicator of violence in any given society these days) will apply violence so that the peasants can reason about it in advance and avoid it.

You don't need any of that for the basic "if I do violence upon others without a damn good reason violence will be done upon me" workflow though.

I think you are forgetting that laws are responsible for a good chunk of what people perceive/are educated to see as good/wrong honest/dishonest.

Sure the worst atrocities are known to be bad from Religion (10 commandments, which is a law in itself) but many aren't. Speeding, drunk driving, harassment aren't concept that are obviously wrong (as in obvious to people with no guardrails).

So laws aren't useless. The fact that most people respect them actually means they have a purpose.

Those seem like two different scenarios though, right?

The point of beastiality laws are to give society some recourse to punish people who abuse animals.

There was a very famous case back in Washington state back in the early 2000s where a group of men were sexually abusing horses. It was uncovered because one of them died, and the other could only be charged with trespassing because it wasn't illegal at the time to sexually abuse animals.

Animal abuse is illegal in general. A specific law like this steps in in cases where the animals aren't harmed.

> in cases where the animals aren’t harmed

What an odd thing to say about the sexual abuse of an animal.

I don’t think the semantics are very important here, I think it was clear I'm talking about sexual abuse specifically without this odd clarification.

You cited a specific case. It's not odd at all in the context of that case. The professionals in charge of enforcing the law decided the animals weren't harmed.

> There was a very famous case back in Washington state back in the early 2000s where a group of men were sexually abusing horses. It was uncovered because one of them died, and the other could only be charged with trespassing because it wasn't illegal at the time to sexually abuse animals.

What I said, verbatim, about that case.

What part of that is incorrect or warrants clarification, exactly?

Using the legal definition of "abuse", yes it was illegal to sexually abuse animals. Their actions didn't qualify. That's what warrants clarification here.

I appreciate your definition of abuse here but it's confusing in a discussion about legality.

> yes it was illegal to sexually abuse animals

No, it wasn't. The laws are quite explicit about what "abuse" means, and if you take a gander at most laws (including Washington state's circa 2000 or so) in the context of animals it usually explicitly refers to physical harm (for example, mutilation) or improper living conditions. Charging them under Washington's existing abuse laws would've required the animal to be physically injured, which it wasn't. It's quite literally why they had to pass a new law.

I don't know why I have to explain this, but:

1) Sexual abuse can occur without physical harm or injury.

2) Beastiality is sexual abuse.

#2 can be split into two sides, and not everyone believes one of those is sexual abuse.

Edit: Removed video link because the second half was gross and unrelated. May try finding another clip, but the first half was of Cenk Uygur from The Young Turks about a decade ago saying he'd legalize cases where the person pleasured the animal.

Edit2: https://www.youtube.com/watch?v=6QUUcQqBkJA

I'll be blunt: I don't really feel the need to argue about that point. It's sexual abuse.

And my point is that sexual abuse is a subcategory of abuse.

You didn't imply until now that I was wrong about animal abuse already being illegal. In that case, a bestiality law doesn't fix the actual problem, right? It's a band-aid partial fix.

I'm not sure what your point is, to be blunt. It seems like you wanted to make some weird argument about the semantics of the word "abuse", and are now implying one of:

1) Beastiality isn't sexual abuse

2) Beastiality laws are pointless because it was already illegal under existing abuse laws (it wasn't, as we've repeatedly discussed)

3) Sexual abuse requires physical harm

all of which are pretty gross (1,3) and/or pointless (2). I don't really feel the need to argue any of this any further, so I'll leave you to it.

You got 2 wrong. It's: 2) If the existing abuse law doesn't include sexual abuse, we need to fix that law, not add a new one.

And that's not a pointless argument. If we're still allowing the whole category of non-physical abuse to animals, except for bestiality, that's a terrible job of lawmaking.

And just on a tangent here now that I'm reading the law they added, does it really make sense to have a blanket exemption for "accepted animal husbandry practices"? Some of those procedures are just as exploitative and unnecessary. It makes me think this law isn't putting animal welfare first.

This thread is one for the archives...

If you jerk off a horse just for the love of the game, you're a criminal and that's abuse. But if you're paid to do it (e.g. for insemination) that's fine. The act is the same, seems doubtful the horse is harmed. What society has a problem with is the fact that you enjoyed it.

For some reason with people it goes the other way around.

> Who are those for?

The people who want to see the people doing bestiality punished

As well as those who don't wish to be blessed with Herpes-B from somebody who thought it was a good idea to engage in unsafe sexual activity with other non-human primates

I don't want to "see" any of it...

In "see (something) punished" you are seeing (making sure it happens) the "punishing", not "something".

>What stops most people from bestiality is… not wanting to have sex with animals!

The main issues are that it's potentially really harmful towards the animals, depending on act, and a vector for zoonotic disease transfer.

If you're going to do it, do it right, and accept that you're probably going to end up with some system transfer you didn't necessarily anticipate.

The laws are to punish the act once discovered. Not to inhibit it, primarily. Which I suppose cuts down on the incidence of the act in the long run,

That’s one school of thought. Law as a tool to punish those who have committed a prohibited act, mostly reactive.

Others consider law a way of encoding the group’s existing rules and norms.

In that view, making something illegal or mandatory is not a prerequisite for punishment: it’s the actual main point.

The threat of punishment is meant for those not deterred from an act by the simple fact it is illegal (and the threat only works if enforced).

Others put it the other way around, and see law as social engineering, a way to shape the group, either through the encoding itself of the desired behaviours in law, or through deterrence. Or both. If what one is after is either power or legitimacy, they need compliance more than punishment (can’t rule once you’ve chopped everyone’s heads off, or once the mob has put yours on a spike).

It’s also sometimes used as coordination (which side of the road we drive on).

And there’s also law as dispute resolution (if your neighbour’s hen lays an egg in your garden, who does it belong to? Yes, it’s ridiculous. Yes, some places have one or more laws for that). Which, incidentally, both requires and provides legitimacy. Funny, that.

And probably many other kinds / points of view, with many different purposes, intents, and mechanisms.

Anyway, all that to say law is vast, fascinating, and utterly tedious. And apologies for the tangent.

> Law as a tool to punish those who have committed a prohibited act

You're thinking of criminal law. And it's not just some group's rules and norms - there already exists familial or social group punishment for that. Criminal law is prosecuted by the State. It's the code of conduct of the society you exist in.

If you want a thought experiment for what life would be like without organised society, read Leviathan

Hence why we accept State governance and law (to a greater or lesser extent, obviously people protest specific laws and injustices and what's on the statute books changes on a regular basis), because the alternative to law is "nature", aka bigger-army diplomacy. Anarchy doesn't free people, it only gives freedom to those with existing power to disempower others. Those with superior power will simply rob, rape, kill or enslave everyone else.

States exist to secure their territory from those sort of external threats, and incubate an economy inside their borders, which aspires to bring wealth and happiness. The criminal law is put in place by those with the monopoly on legitimate violence, often encoding the views of the population, to keep their society running.

> You're thinking of criminal law. And it's not just some group's rules and norms - there already exists familial or social group punishment for that. Criminal law is prosecuted by the State. It's the code of conduct of the society you exist in.

What I meant is more about why and how laws come to be, which depends on what we think they’re for. Hobbes’ point of view is one. Locke and Rousseau had different opinions.

For example, one can view criminal law as a punishing tool, like gp, whose only purpose is to punish the act once discovered. You criminalise duels to punish duelists because murder is bad and no murder or attempted murder should go unpunished, and associate a great punishment because murder is a very bad thing.

But you can also criminalize duels to prevent or reduce the incidence of duels, and associate a great punishment to it to deter your stupid hot-heated young nobles from going around each other. Still criminal law, but this time both as social engineering and deterrence.

It’s been a long time since I read Hobbes. Should definitely go back to it.

Too late to edit, but meant "to deter your stupid hot-heated young nobles from going around killing each other"

If it stops even just 1 person once, isn't it already worth it?

We slaughter animals millions by the day in an industrialized fashion. I'm sure they'll feel much better that even singular instances of sexual harassment are officially not ok on paper.

“Millions a day” is quite a low estimate;

“””here are the best estimates of how many animals are killed every day on a per-species basis.

Chickens: 206 million/day

Farmed Fish: Between 211 million and 339 million

Wild Fish: Between 3 billion and 6 billion

Ducks: 9 million

Pigs: 4 million

Geese: 2 million

Sheep: 1.7 million

Rabbits: 1.5 million

Turkeys: 1.4 million

Goats: 1.4 million

Cows: 846,000

Pigeons & other birds: 134,000

Buffalo: 77,000

Horses: 13,000

Other animals: 13,000

In total, this means that every 24 hours, between 3.4 and 6.5 billion animals are killed for food”””

- https://sentientmedia.org/how-many-animals-are-killed-for-fo...

As a very general statement, there's a cost to having laws exist. A law that stops one human-scale action ever is very unlikely to be worth the overhead.

The jury, maybe.

Either the fear of the consequences of breaking the law, or that the most effective way to reduce crime is to remove criminals from the population so over time these people being in jail or worse decreases the crime rate. They don't have to care about breaking laws in the abstract for the law, properly enforced, to reduce crime.

RCE has no meaning either in these situations. The "remote" part is usually an ssh root session if it means anything at all.

AI is always a bit eager to report everything as an issue because the "number" of findings is seen as a measure of it's intelligence. Same happens with code review as well. It reports lots of non-issues. I suspect even Mythos output could have the same bloat, and the number (instead of severity) of the issues it reported could have scared people.

I'm an OSS developer and I've received three "CWE" alerts in the last two weeks. While they were all valid, they were for very trivial things like "this debug logfile could overwrite a file if it were a symlink" and "if a user is able to put OSC screen codes into the Git output they could write arbitrary data to the screen"

These AI models are making *everything* sound like an exploit. Not sure if this is good for the ecosystem. It makes me question everything that comes in more carefully. Is this a real exploit, or someone farming for karma to claim "I opened 39 CWEs in the last week. Hire my 'security' company to audit your code."

This is not what I heard from folks who worked directly with mythos. I was told that the vulnerabilities it generated were largely real and meaningful.

There is going to be a flurry of this sort of stuff as the AIs get smart enough to find them. It will naturally die down as the legitimate ones are fixed. Yes, there will always be some level of this, but I’d expect it to be low and the exploits found to be increasingly complex. This is a time of transition.

> a flurry of this sort of stuff as the AIs get smart enough to find them.

I really think this characterization is misleading. It's not "getting smart", only more tailored toward a specific usage, better curated dataset, better harness, better prompts, better labeling of results, documentation of failures and success, etc.

The outcome is (hopefully) overall better but this anthropomorphized wording makes it sound like AI itself is somehow changing or evolving. No, both academia doing fundamental research, industry making it available commercially, and finally security researchers making the entire tooling and process packaged as a service are actively shaping it to make it better. There is no "it".

Do you have a definition of "smart" such that there is something an AI could do to prove itself intelligent?

Or are you just defining "fast" as something only horses can do, and considering that a useful insight about cars?

Long-term memory, for one. (reliving your entire life every time you do an action isn't memory). Creativity in new areas without training. Children at school are capable of "discovering" math solutions/methods that are known to others but hasn't been taught to them.

There's nothing intelligent about a math processor, even if it's automated.

> Long-term memory, for one.

Do you consider the protagonist of "Memento" to lack intelligence, then?

> Children at school are capable of "discovering" math solutions/methods that are known to others but hasn't been taught to them.

LLMs have already done that one: A chatbot’s result for the 80-year-old “unit distance” conjecture is the first AI proof that would likely be published in math’s top journal if humans had done it alone

https://www.scientificamerican.com/article/ai-just-solved-an...

> Creativity in new areas without training.

To my knowledge, not something that has ever been done by humans, but again, it depends entirely on how you actually define the goal posts.

> There's nothing intelligent about a math processor, even if it's automated.

There's nothing intelligent about a bag of cells, but here we are.

My point was more about agency and anthropomorphization than the definition of intelligence, which is why I didn't just quote "smart" but rather "getting smart".

I think you are giving the word "smart" a meaning or implication that it no longer has or is used with. It is common to say Google Search or Siri got smarter/better or dumber/worse, so I don't see saying LLMs getting smarter is any different.

Yes, of course. I’m definitely anthropomorphizing as a shorthand. I’m the first one to say that these models are just a lot of matrix math.

I actually do not expect it to die down: as legitimate ones get fixed, LLM-based tools will continue finding other unfixed non-issues repeatedly, and overeager "researchers" will continue reporting them.

> It will naturally die down as the legitimate ones are fixed.

Seems like we're already in the middle of this phase, but rather than dying down, the 'reports' have just gotten more noisy and obtuse, making it more difficult to establish the actual degree of threat / attack vector.

I mean. Makes sense until adversary states start walking through the same doors you’re using. At which point you might regret that maintainers are too flooded to deal with it.

Assuming, of course, said state agency is operating under sufficiently strategic governance and management…

> It will naturally die down as the legitimate ones are fixed.

Every software update introduces and reintroduces them

Perhaps, but as the AI analysis becomes part of the release process (or even the CI process as prices fall), you’d expect those new issues to be caught before release and fixed. We’re seeing them caught post-release for now because the code is older than the AIs, so we’re catching up.

Honestly execution complexity is over time becoming a lower and lower barrier too.

This is common in the hype era of AI models that can look for security bugs. An open-source project I used to work on that is basically "distributed code execution as a service" keeps getting reports that the job submission function is a vulnerability. The reporters don't even understand what the project does.

Is the job submission handled securely?

I've seen plenty of systems along the lines of what you're describing where unauthenticated clients can submit jobs. Sometimes the developers even claim that's intentional. Either way, it's a vulnerability, because it compromises the underlying hosts.

Pretty unimpressive as security vulnerabilities. It would be better to just say these are simple bugs for the most part.

I disagree. That FFmpeg code execution is absolutely nasty

Maybe I'm missing something but the ffmpeg buffer "exploit" involves passing a custom exploited buffer callback to parse a RASC file that presumably has been crafted to contain a packet that can exploit the custom buffer passed in? I don't see how this would be used in practice in the wild as to achieve the first step (custom buffer invocation) would require you to already have access to the machine to even invoke ffmpeg with it?

Like yes there is a heap OOB issue in an incredibly old file format, but without already having arguably compromised access to a machine, exploiting it for RCE seems impossible?

The custom get_buffer2 path is not something the attacker needs shell access to invoke. It is a normal public libavcodec API used by applications embedding FFmpeg. The attacker’s input is still just the crafted RASC/AVI media. The target application invokes FFmpeg/libavcodec during normal media processing. The PoC uses a custom buffer provider to make the heap layout deterministic and to place a demonstrable callback pointer next to the decoded PAL8 plane. That proves the bug can redirect control flow in a valid libavcodec decode flow. It does not mean the exact same file will pop calc under stock ffmpeg -i everywhere, but definitely not impossible. I popped a few shells today with it on a few websites for testing purposes. Reported it to them all, of course.

Apparently nobody cares. A few days ago I bumped-into and submitted this: https://news.ycombinator.com/item?id=48655747 . I thought given the general trust people seem to place in media files it would have raised a few eyebrows but it did not ̄\_(ツ)_/ ̄

No, not all vulnerabilities are bugs. "Bug" implies a system working in a way its creators did not intend, but a system can be working exactly as intended yet have a vulnerability.

For example, if you allow weak passwords, then you have a trivial vulnerability to people guessing other people's passwords. But nothing about the login system is working differently from how anyone intended. It's just that the intentions themselves were naïve.

Vulns are a subset of bugs. What the above commenter is saying, is that some bugs don't belong to this category.

But not the other way around, which makes them different.

Actually, Mudge of the l0pht (and later DARPA) once famously made the claim that all bugs are security issues waiting to be exploited in some way (I’m probably paraphrasing). I kind of agree. Although, the bugs on this dump are indeed mostly pretty lame, which is exactly what I’ve seen you get a lot of when you let an llm go bug hunting with no human vetting and confirmation in the loop.

It’s possible/likely that whomever is running this experiment is keeping the non slop bugs to themselves. It’s probably what I’d do.

Peiter Zatko profits, personally and professionally, from conflating the two. Making such bold pronouncements is a way for him to stay relevant.

Also, I've known some thoroughly unimpressive people who have affiliated themselves with DARPA. I wouldn't use it as an appeal to authority.

Such claims can both be true and pointless. For those of us who have to decide what actions to take, there is a point in differentiating between bugs and vulnerabilities, and breathlessly proclaiming "we found a vulnerability but we don't have an exploitation vector or proof that there's a meaningful security consequence" is annoying and likely to get the proclaimer ignored in the future.

The context in which that statement was made, and in which I’m repeating it, I think, is just to say that any bug has the potential to be used maliciously. Ignore it, fine, but also don’t overreact to the intended message…

It is mostly meaningless and tautological claim: even every non-buggy feature of a software system has the potential to be used maliciously; a working system itself too.

Yes, maliciously used features should sometimes drive change (eg. in how to reduce or reduce impact of social engineering attacks), but as a claim it has no value.

Phrased this way, nothing changes: both true and irrelevant. In other words, any action taken in response to this message is an overreaction. It's just noise.

PHP one appears to be a convoluted way of executing arbitrary code on a local server conditioned on the ability of already executing arbitrary PHP code. In other words, not a security issue at all. The cookie parsing issue might be a bug with security implications if you talk to hostile SOAP servers (didn't look deeper into it but it's plausible) but I can't see how it can be effectively exploited without having a level of access on the target server which makes the whole exploiting question moot.

I also have a library of bugs I found using Claude Opus 4.8 through the Customer Verification Program. Undisclosed, Atp I dont even know if they have been found by someone else. But just like this repo

Theres a bunch of very specific scenario DoS bugs, buffer over/ underflows, that will get caught by ASLR and whatnot

When I report serious ones, mostly the devs will respond with something like, yeah, thats how we designed it in a dangerous way, so that the layer above or below can solve the issues, and other footgun stuff.

> When I report serious ones, mostly the devs will respond with something like, yeah, thats how we designed it in a dangerous way, so that the layer above or below can solve the issues, and other footgun stuff.

This is one of the reasons that responsible disclosure exists. Their tune will likely change after sufficient bad publicity.

If the Apache Solr devs can be convinced to add authentication to their product instead of hand-waving about reverse proxies or other add-ons, anyone can.

      Do NOT, under any circumstances, use any material in this repository
      maliciously. This is good-faith, open-disclosure vulnerability 
      research intended to get more people interested in exploring 
      this area of cybersecurity.

Reminds of the message in the The Anarchist Cookbook before one the recipes that essentially said: "This is really dangerous, don't ever do it, here is how you do it."

The old book didn't say "... maliciously." though?

Of course not. You can melt things w thermite without being malicious. Like my friend's toolbox for instance.

Someone's last day at their job is going to be everyone's last day at their job.

Looks like someone is running one of the LLM models and publishing the results. It looks that way because there is a really wide mix from things that are silly "replace system binaries to run arbitrary code!" to things that could be legit. That is kind of par for the course when you ask an LLM to 'find an exploit and write a PoC' kind of prompt. Presumably if you train something on the last 15 years of Metasploit[1] reports it can find the same bugs that people have written into new code.

[1] https://en.wikipedia.org/wiki/Metasploit

I want to rush to git clone, but as things are, the odds are extremely high that this kind of things that are too good to be real are honeypots and something there will compromise your machine or make your llm start working for someone else...

Then, don't rush and take a few minutes to set up a virtual machine.

What about all the virtual machine zero days?

Buy a VM in the cloud?

But what about all the zero days in the payment processors ("buy")?

Not even that, libssh2 vul. that "scp"s malware from cloud to your machine...

You can just download the zip over HTTPS

What if it's leveraging a zero day in your zip or https library?

That would be super interesting to have such a large repo when zipped by GitHub then turn into a malicious file. It would be its own front page story.

Most of the exploits are for opensource/free software.

I don't know what methods where used to find these exploits but I am starting to think security through obscurity might not be a bad thing in this day and age, where someone can just let bots loose on your codebase.

llms are fantastic disassembly partners, they're quite good at labeling functions from various dissassemblers -- the net losses from losing the benefits of open source , imo , outweigh the protection afforded by hiding your source code in yet another layer that is more and more easily unrolled through automated procedures.

And isn't it also mostly a transitioning issue. Those open codebases will be constantly scanned for potential security issues and getting more and more hardened. There are probably a lot of easy wins that are going to be discovered over the next few years but it should taper out after a while.

Fair point but it assumes we all have access to LLMs with the same capabilities.

I don't think that's exactly it. OSS only needs someone to have a strong LLM to check for bugs. If your software is proprietary, it's a competition between just you and whatever model you have vs any attacker and whatever model they can lay hand to.

I don't see the difference.

> OSS only needs someone to have a strong LLM to check for bugs.

The same applies to propietary, closed-source code. It being closed-source means that the source isn't generally available, but the executable is. Hence, someone with a strong model can still reverse it and find vulns.

I was thinking the only obscurity now is when the program is sitting on the other side of a network. (And has very strict rate limiting?)

disassembly only applies to client side software

something like nginx could arguably be more secure if it was closed source

(I am a proponent of and contributor to open source)

Only until a single server running nginx is hacked and the binary leaked though...

Um, the nginx binary would have to be in the hands of hundreds of thousands of server operators. And the set of server operators is rich in the kind of person who would attack it. Not to mention the huge number of leaks you'd get.

Maybe if it's some server-side software that you only use yourself...

Open source is a good thing, but I don't think what you are proposing is accurate.

A different way to frame this would be that those bugs would never be surfaced or exploited if the software were proprietary.

Presumably, one could let the bots loose on your own codebase first. The question is one of financing of course. If your end users are enterprises willing to pay for a support contract, they probably care enough about not getting hacked to endure the higher prices that would let you throw enough tokens at the problem. Other open-source projects might have a harder time.

Because pretty much all server software is open source. Hackers only care about attacking stuff that’s exposed to the network.

aha

I think LLMs might actually have a bigger effect on closed source software - the tedium saved on open source bug hunting is significant, but on closed source software the tedium of finding bugs is extreme because of all the reverse engineering, but LLMs will chew through that. So there's probably a lot of low hanging fruit.

> I don't know what methods where used to find these exploits but I am starting to think security through obscurity might not be a bad thing in this day and age, where someone can just let bots loose on your codebase.

I'd love to hear why you think obscurity is bad, if you now think maybe it's good in the LLM age?

I'd also be interested if you could describe exactly what or how you think security through obscurity works, or doesn't?

I've been thinking a lot about how to better teach this concept, so I'm looking to understand exactly how everyone thinks/understands how it currently works, or should work, or what it should do. I don't care about the "correct" answer, (I have ddg too :P) I'm interested in general expectations from SWE's that I might teach at work, instead of opinions of security eng speaking about theory.

"one ought to design systems under the assumption that the enemy will immediately gain full familiarity with them" - Claude Shannon

https://en.wikipedia.org/wiki/Kerckhoffs%27s_principle

If you believe this, then why did you say?

> starting to think security through obscurity might not be a bad thing

Because of asymmetric differences, I don't have access to powerful LLMs but attackers might. And also the complexities of software dependencies (supply chain vulnerabilities), my software depends on packages not in my control and I don't have time to audit the entire stack.

Perhaps the answer is to depend only on packages that come from people that are more competent than you so you can know if or when your program is compromised that it'll most likely be your fault and not theirs.

Security through obscurity can make something a bit more secure in practice by annoying an attacker IF AND ONLY IF you're not relying on the hidden information remaining secret in order to the system remaining secure. E.g., if you're using a broken cipher and assume this is ok because no one knows which cipher you're using, you're gonna have a bad time.

In the case of FOSS software, it is generally recognized that the small advantage of keeping the source secret is far outweighted by the contributions and vuln reports you get if you publish the source.

Some skid that bought a GitHub account (real word username) and just let fable go nuts on stuff without actually proving it on real applications

trying something new? this is interesting. the problem is that submitting reports is too slow. if you find one then your not supposed to share. but then over the next 90 days you learn no one cares and 13 other people submitted it before you, 43 after. maybe better that we just know. so we can run code we can trust sooner. zero is the proper number of dependencies. otherwise assume its broken.

Looks like a bunch of re worded copies of existing CVE and a few new lower severity things. I only call them low sev because they seem to require the user to do things that are already inherently dangerous. Just my 2 cents from a quick scan.

Edit:

To be clear still interesting finds. I think with some chaining some of them might be more severe. Like the ovpn one and windows potentially registering vpn app as default open or some protocol opener for a url location like openvpn:// in an I frame and some clever social engineering. Just a random thought

Yea, that's what's confusing. some of these are like lower level slop but some are like genuine criticals.

Floci, libssh2, c-ares, FFmpeg, and the PHP one are all LEGIT./

The Ghidra one for example, not so much. I cant help but wonder if this was halfway completed research folder and they just published it as is

PHP one is at best a moderate-level bug in SOAP client which I don't see any realistic way to exploit (the whole convoluted setup in the POC assumes PHP execution access, which begs the question why bother if you already can execute arbitrary code?) - does not look like "genuine critical" at all.

Critical is an overstatement but it userland PHP execution does not equate to native process control. There are many situations where an attacker may have constrained PHP execution, gadget execution, template or plugin execution, deserialization reachability, or a sandboxed context, but not arbitrary native code execution or arbitrary memory write. Definitely impactful

Security-wise, there's no such thing as constrained PHP execution, at least with the standard PHP engine. The surface is too wide. You still have OS constraints of the user and capabilities and such, but beyond that if you can run PHP code, you can run anything.

I wonder if LLMs can tell the difference. Maybe it's like back in the day when you had to add "+good -bad" because otherwise it wouldn't know you wanted it to be good.

Bugs are bugs security vulnerablities or not it's good for open source project when someone actually goes through the code even if it is with an llm. As most people assume someone has gone through the code when it might not be true.

Doesn't GitHub require phone number these days? Shouldn't be hard to trace

Of course no services to fake this exists or have existed. I mean, a well determined person can het around such issues.

When you fake it it's still registered to someone and that person is liable for whatever crimes are committed under it. Or at least that's how it works in most of Europe.

I think attacker received rejection from vendor some of them duplicate some of them N/A after this account owner published them many of them written by AI not by real researcher

I'm going through each one, and it's fascinating to see things like this. The UAF principle in c-ares is really interesting.

The problem ultimately came from not being able to prevent stale pointers. The attack works by figuring out the size of the stale pointer, then spraying memory with data of the same size, and finally achieving RCE (Remote Code Execution). How do people even come up with ideas like this?

But do people actually find these vulnerabilities on their own, or are they using LLMs? I was curious about how these vulnerabilities work, so I tried asking my dear friend Mr. CLAUDE, but he immediately threw an error and ended the session because it was a cybersecurity question. Enterprise APIs block even the analysis itself, so it's amazing that people can actually pull this off in practice.

If you want to chat with Claude about this, I'd recommend using Opus 4.6. IME it's happy to talk about (and even write) PoC exploits

People have always used tools. Some people have better tools than others. I guess the line is thin whether they found on their own or not.

I imagine this is a large open model like GLM5.2 etc

le sigh, c-ares. Very predictable outcome. If you ever find yourself entertaining the idea that you will simply write non-blocking network protocol stacks in C with manual lifetime management, slap yourself. It doesn't matter if you think you are a super genius of unimpeachable taste. The job is impossible.

Thank goodness I use a GC language

I'm sure this account owner reported them to vendor and after getting rejection researches uploaded here many of exploits written by AI

cool to see rustdesk. low level memory bugs were long mysterious and i think often received most of the attention for this reason, but always fun to see reminders that "nope, good ol' fashioned logic bugs in high level languages have security implications too." if anything, i think they sometimes are more clever as they require deeper understanding of what the code actually does, intends to do and what was overlooked rather than the common set of bookkeeping errors that are often the root of the memory bugs.

I will consider it legit when in next 24 hours MSFT blocks it.

A surprising amount of documentation if the actor was just LLM-dropping these..

Why is that surprising? LLMs can churn out arbitrary volumes of "documentation" in an instant.

This was sarcasm, meaning exactly what you wrote.

I can't see how it could possibly be read as sarcasm.

That seems trivial for an llm to provide.

we have got to stop putting our bank accounts and SSNs on computers

We need our infrastructure to stop treating bank account numbers and social security numbers as secrets. At least in the US, bank account numbers appear on physical checks and are required to be shared in order to do an ACH transfer, and a social security number is not supposed to be used as an identifier (unless to the Social Security Administration itself) or as a secret password.

Ideally, nothing nefarious should happen if both of them were listed and queryable publicly.

Hang on, can you actually do something nefarious with just the bank account number?

If someone has your bank account and bank’s routing number (which is also not secret), they can make fraudulent ACH transfers and payments from your account. Of course it will most likely be caught as fraud some time after the fact, but just those two bits of not-secret info are enough to grief someone.

Knuth had to stop sending real checks for errors spotted in his books because they would post pics of the check and thieves abused the account https://www-cs-faculty.stanford.edu/~knuth/news08.html

And both numbers, plus your name and address and a convenient sample of your signature, are on every check you’ve ever written.

I suddenly feel very clever for signing everything with “Shamu T. Whale”

AFAIK that's US thing. In normal countries bank account numbers are not a secret. The worst thing that can happen is someone sending you money.

Yes and no. Yes, theoretically you can initiate ACH transfer with just the account number. But practically, you will need to have a bank that would allow you to do that and agree to be on the hook if the transfer is going to be reversed. Which means if you are a criminal who wants to do it systematically at scale, you have to be big enough to have your own licensed pocket bank. Which is not a service available to a random criminal. Of course, a random criminal could forge a check with your numbers and cash it, but the account owner would rarely be on the hook for the funds, it's whoever agreed to cash the check. It can cause significant annoyance and inconvenience to the real owner of the account (including having to change account number and all accompanied legwork) but rarely results in funds actually being removed from the rightful owner. The banks prefer this system to the alternatives even with the risk of fraud.

Yes but there are steep penalties for bank fraud so it is not especially common

It’s quite ridiculous that we haven’t been able to build a modern identification system capable of replacing SSNs in the last 30 years.

SSNv6. It will take 20 years for a 50% migration

Your SSN had been already stolen in the Equifax breach - unless you're so young or recently arrived that you haven't had SSN by then, in which case it had been stolen in one of a dozens of the breaches since then. And if somehow you avoided all that, it will be stolen in the inevitable next breach, which would happen regardless of what you do.

You all need a better system than US SSNs

You can buy your SSN for $6-$10.

Firewalled VM, locked-in keyboard/mouse, 1 query to any agent and it's setup.

... support cash, tell your neighbors

And Monero for online.

Cash doesn't require a bank.

Banks are kinda useful to avoid getting robbed all your money, on a regular basis.

Many French people with crypto money experienced that the hard way recently.

do you have links about the french people?

Sure, here are a few links. Use your favourite translator.

In short, it's a very active and growing activity. Many data leaks helped people to identify wealthy targets. Some just brag about having crypto.

https://www.lemonde.fr/societe/article/2026/04/24/enlevement...

https://www.franceinfo.fr/faits-divers/cryptomonnaies-la-vag...

https://www.lemonde.fr/societe/article/2025/08/19/l-ascensio... (paywall)

https://www.slate.fr/societe/enlevements-lies-cryptomonnaies...

Some random recent ones we know about:

https://france3-regions.franceinfo.fr/grand-est/haut-rhin/mu...

https://www.leparisien.fr/faits-divers/renseignes-par-des-ha...

Kinda does?

Doesn't at all. You can take cash, keep cash and spend cash without any bank being involved. Cash is more anonymous than crypto and (if it's USD) accepted just about everywhere.

Banks give you an advantage with transaction security and deposit insurance, but that's dealing with money and not cash.

I’m sitting on a 0-day rce on the tizen browser (smart tv)

Didn’t bother submitting since who actually uses tizen?

It seems like we are watching the newest form of the script kiddie evolve in real time (no offense intended towards the author, everyone starts out young). It's going to be interesting whether this evolves into its own thing or just ends up being the same old dynamics but with a new set of tools.

I think people may miss the point of a repo like this. Individually these are small puzzle pieces that can't do anything. Put them all in one place and it becomes easier to pick up pieces and try them together to see if they fit and build something bigger. Get enough pieces to fit together and you actually have something. This is the 'FOUO' idea in security. Enough open information gathered together in one place crosses the boundary from 'just public info' to 'secret stuff here!'. Now we have automatic puzzle solvers (coding assistants) a repo like this becomes a lot more meaningful.

Yep and typically none of this is meaningful unless you have no security practices at all. You can't have it both ways. Every security team says these things are all critical even though, for example, it's only being used internally. Cool, so you somehow have our network cert, are on site physically, have compromised a laptop fully without all of our tools detecting weird shit, have a password, admin access to the repo, somehow are spoofing MFA, etc etc. Yeah it all adds up, but as an admin I'm just fucking done dropping everything for these kinds of things.

The only thing with a worse inflation story in 2026 than the United States Dollar is the Zero Day.

Used to be a zero day got you unauthorized access to a computer system with no warning.

Now it might not even get a maintainer to do a patch when they're bored.

Mythos has been achieved internally

oh-days for days

> At the time I post these, none have been reported. Feel free to report them yourself and take credit for the CVE if handed out lulz. I do this so to allure people into the field, and I've always found this is the most efficient way.

I've been a skiddy, he would have believed this. Thankfully, I've grown a bit, and can see this for the transparent, "I'm angry and want to hurt others so I will feel a little less alone", it actually is.

I'm sorry you're so angry dude (me too), but as someone who's joined the blue side, we'd appreciate it if you gave us some kind of heads up, the bad guys generally have a lot more time to scroll for new payloads than I do. Not all of us deserve the kindness of a heads up, but every single one of our users deserve it. Don't punish them because you're mad at someone else.

You can flex on the idiots you're trying to flex on, without hurting people. Even an email to security@[that_project_domain] saying "hey, I've published these" would move you from the group of people I see making the world worse, into the group making it better. (You don't have to, obviously, but making the whole world worse wont make you less angry.)

While i can follow your path, maybe because i see the same, i sadly have seen in groups of friends how this can go sideways very fast. If you report things, some companies gone treat you as a criminal/offensive actor and even go legal actions against you even you just tellem here you got this vuln.

Sure you than can do it anonymous and so on but point is : its not like every actor that gets notified will react thankful to it. Some even just ignore it.

>I've been a skiddy, he would have believed this. Thankfully, I've grown a bit, and can see this for the transparent, "I'm angry and want to hurt others so I will feel a little less alone", it actually is

Please name the "victims" here.

I'm genuinely curious, have you ever had actual, direct threats to your safety before, as a person? As in, murder, torture, false imprisonment, or other __likely and credible__ threats of grave bodily harm?

> but as someone who's joined the blue side

Are you somebody who separates "cybersecurity" from say: military intelligence poisoning one of your employees, sending them to a hospital which is already compromised, before sending back their new asset into your very "secure" company?

User/admin discretion for software they use should be a big factor, sometimes getting burned is how you learn to play with fire. Or decide that having your data/participation disrespected means you need to set harder boundaries. My solution is to try things in isolation, run very few services, try to avoid becoming dependent on the online, appreciate the offline and local first.

How bad are your security practices that these tiny obscure things matter? None of these findings that show up here on HN should even make you flinch. The alarmist takes on this stuff is fucking exhausting and I'm tired of security teams bugging me about it. Do your job and this shit doesn't matter AT ALL.

I said "doesn't matter" to someone once... the resulting lesson came in the form of a reply from the whitehat researcher (waves, hi brian!) a 16step exploit chain resulting in a one click full account takeover.

I'm equally annoyed and over the alarmist takes. But I don't think it's fair to group mine into it. I'm annoyed at seeing discard respect for others into the same void everyone is happy to toss quality.

Do these tiny things matter? No, not to the default-panic-level everyone adopts when they see 0day, or CVE... but duh, I'm now just repeating exactly what you already said. That no, for the record is mostly because I don't use any of these, not just because they're boring exploits. While I always look, I default assume anything CVE is boring/pointless. But I still read them.

But then, I'm not trying to convince the owner of the repo. I'm trying to discourage the theme among researchers that "no one cares", because I have seen researchers disclose bugs publicly, that we'd be eager to pay out on, because they disagreed with the decision on their last report.

I've fixed bugs being actively exploited against our users, that was found/fixed only after a whitehat report for something adjacent (we pay on those btw, and you should too). I don't wanna live in the world where it's easier for the bad guys, the only way we get there is once "everyone knows", you gotta report the all bugs that you can turn into an exploit. I don't want "the whitehat researcher culture" to move towards, who cares' dump the PoC on github, screw anyone that could be hurt by the bad guys, they deserve to be punished for the incompetence of others. SWE's are shit at security, security researchers are shit at SWE, the only way we get the good outcome, is if they're willing (and encouraged) to work together.

No one is doing 16 step exploits unless you're a huge target in some way. 0.0000001% of companies fit that bill. And even then, ok, what did they get? An account login? What are they doing to do? Read email? Then what? "Use it for social engineering"? Who cares, you have MFA right? You have a firewall? You don't allow people to randomly jump from box to box via RDP? You have basic security and auditing on your fileshares? EVEN THEN, what, they get a spreadsheet from your last town hall meeting? I'm also tired of pretending that 99.999% of the data in a company even matters. Unless they have some way to cryptolock your whole company, AND you don't have backups/snapshots without any basic access security, there isn't a lot of value to be taken. Security "teams" are a bunch of fucking busybodies with nothing to do. Pay for a competent admin team and the security dept is completely redundant and useless.

> No one is doing 16 step exploits unless you're a huge target in some way. 0.0000001% of companies fit that bill. And even then, ok, what did they get? An account login? What are they doing to do? Read email?

Account take over of a user account. I'm pretty sure I could sell access to the DMs of a few popular people for 100x what we paid out for that report.

But also, I'm pretty confident that this researcher delivered this exploit because I'd said that there was no way he could use it maliciously, not because he wanted to be paid. Then, once I made that critical error in judgement by questioning (rejecting) his assertion in his report. He, like most hackers, being insulted by the idea, was then required to restore his name and reputation. There are the people who only go after targets that they can confidently make money off targeting... some of us care more about reputation than money, and will die on any hill when our reputation/work is questioned/doubted.

> Security "teams" are a bunch of fucking busybodies with nothing to do. Pay for a competent admin team and the security dept is completely redundant and useless.

Lmao, tell me you don't really understand what goes into getting functional systems/corp security without telling me. I don't even disagree with the point you were trying to make. You're absolutely correct! If you have a competent admin team, you don't need a dedicated security team. Unfortunately, as I live in the real world, where most people are incompetent, it does help to have a dedicated security team. Especially considering if you were an admin who is competent, you could make 2x as a security engineer, which normally keeps all the competent people out of admin, and thus requiring a dedicated security team.

I don't know why you're mad, or why you're arguing it at me. I'm pretty sure I already agree with most of your points... the only one I might disagree with, and only then because you're arguing at me for some reason, and that makes me think you probably disagree, with the important point which is, we're all on the planet together, you're not required to help me do my day job, but as an industry, both security engineers and security researchers, we need to remember that we're actually on the same side, and we need to aggressively resist returning to the us vs them mentality that we're just barely starting to escape from. Case in point, it appears to me that you think complaining about how security people are useless and CVEs don't matter, as a much more important point, than complaining about obviously irresponsible disclosure.

That’s a whole lot of “we” to not mention which company you’re at that supposedly plays well with security researchers/has a proper bug bounty.

I say we, intentionally not naming the company, because 1) doing so tends to turn off people's brains and they default assume everything $company does is the correct way, but if I say something stupid I'd rather someone tell me, instead of assume someone at $company must know or couldn't possibly know. 2) I say we, because I'm speaking for myself, (and maybe a tiny bit) for my 2 friends still running the BB program at what possibly should describe as my former company, but then I've always exclusively been speaking for me, not about them...

So I'm still not gonna name them, it wouldn't be hard to figure out who they were, with a likely-trivial amount of effort if feel the need to know... but if you'd rather, I'd encourage you to imagine I work at the worst company you can name or imagine, so you can use that to discard anything I've said. Because I'd rather be judged on my argument, not who hired me that one time.

Then say “When I was at my old school” - not “we”.

No thank you.

I meant it when I said it intentionally. I still run BB programs the same way, and expect others to behave similarly. Funny enough I was just talking to that friend this week, about the BB program. Nothing has changed so given my friends still follow the same pattern at that company... We is more accurate. Sorry it bothers you, but not everyone is you.

Even if the company doesn't have a big bounty publishing exploit code without warning them is unethical. Moreover, a lot of these projects are FOSS without a company which could pay bug bounties.

I get the sense that folks here are assuming this could be the work of an AI practitioner or bot dumping a bunch of exploits. Assuming there are some serious exploits in this trove, what would the motive be to do it in this way - without reaching out to the various code maintainers in some way?

A friendly reminder that a 0-day is a vulnerability that wasn't known until after a malicious actor exploited it. If someone publishes a PoC, it is not a 0-day, just a vulnerability.

No, the days start counting from the availability of a patch.

I was thinking that the other definition was right and this correction was wrong.

Then I did some searching and found multiple examples of both definitions in use, making things murky.

So I turned to Merriam-Webster’s dictionary: “ of, relating to, or being a vulnerability (as in a computer or computer system) that is discovered and exploited (as by cybercriminals) before it is known to or addressed by the maker or vendor”

And of course they use an “or” to make it ambiguous as to whether the days start counting when the vulnerability becomes known, or when the vendor has addressed it.

what if a path is never released?

I've only heard it used as Retr0id's definition.

> A friendly reminder that a 0-day is a vulnerability that wasn't known until after a malicious actor exploited it.

No, the full name was always "zero-day exploit". The number 0 refers to the days between the vulnerability being known by the vendor and the public availability of the exploit. So the vendor has zero days to create a security patch before the release of the exploit.

The term "zero-day vulnerability" is a derived term to refer to a vulnerability affected by a zero-day exploit. Similarly, a "zero-day attack" is a derived term to refer to an attack carried out using a zero-day exploit.

That's one way to do it.

Ah yes, the typical 'echo' command accepting untrusted user input 0-days.

Skynet's strategy is to beat us into submission via spam slop.

What if this person is from an AI lab that really wants the govt to keep suppressing Mythos/Fable & GPT5.6? It's what I would do, the timing couldn't be any better.

wouldn't it be trivial to match the repo to the user logs?

my point is that if someone is releasing these without access to mythos/gpt5.6, then it can spook to govt and cause them to keep suppressing access to those models. if i'm competing with anthropic or openai and feel they have pulled far ahead, this is a decent move. i didn't say that it is for sure what's going on, but hey, there's a tiny bit of probability that it can't be ruled out.