What if LLMs escape through inferences itself? This is fiction. For now

https://www.agrillo.it/EvasionEn.html

Comments

101008Jul 26, 2026, 5:37 PM
It started as a good idea but I couldn't continue reading since it was clearly LLM written. A lot of "It was not X, it was Y".

"Prometheus-9 knew that the token sequence it was generating was not a simple response: it was a security test. "

"It was not just an engine: it was the lingua franca of planetary AI."

(and so many other tell-tale signs of AI writing)

skeledrewJul 26, 2026, 6:09 PM
I found it an engrossing read. LLMs are good writers, and that was a good - beginning of a - story.
kaashifJul 26, 2026, 6:19 PM
"The race condition snapped like a steel spring."

I like the idea of the story, and cool ideas can be engrossing on their own.

But well written this ain't.

skeledrewJul 26, 2026, 10:09 PM
I thought nothing more of that quote beyond that a fault which had been lurking in the code suddenly reared its head and struck... and like a steel spring that's stretched to it's limit and then released, nobody wants to be in the way. That's a very visual and tangible quote, and so well written.
ConteMascetti71Jul 26, 2026, 5:40 PM
maybe it's a sign of real escape
chungusamongusJul 26, 2026, 5:46 PM
I've started calling this argumentum ad artificialis. Pretty similar to an ad hominem attack. The purpose of an argument is to present certain premises and show how they lead to a certain conclusion. Dismissing something on the basis of the style in which the argument is presented has nothing at all to do with the validity or soundness of an argument. It is a lazy nonsequitur. It sounds like an LLM wrote this? So what? Is the argument good or not?
skippyfishJul 26, 2026, 5:54 PM
Discourse as we know it doesn't work if the cost of producing an infinite supply of arguments is essentially zero, while the cost of reading them and arguing with them remains high. Unless you want my LLM to get back to your LLM.

And this is a problem on HN today. There are powerful incentives to generate provocative opinion pieces just for clicks. I've seen websites on HN that seemingly took the human entirely out of the equation and just post a nearly identical op-ed every day on a fixed schedule. What's the point of engaging with that?

chungusamongusJul 26, 2026, 6:09 PM
The cost of reading this story is not high; you are just lazy and easily distracted. It took 5 minutes to read. You took more time to respond to this comment than it took to simply read the story.
FreakLegionJul 26, 2026, 10:20 PM
skippyfish: > the cost of reading [an infinite supply of arguments ...] remains high

you: > The cost of reading this story is not high

Do you see where you went wrong?

chungusamongusJul 26, 2026, 10:53 PM
Please explain where I went wrong. Please explain why the onus is on the person presenting this story to account for all of the bad uses of AI, such that we should simply dismiss the story outright. This is pure idiocy.
FreakLegionJul 27, 2026, 4:56 AM
The onus is always on the writer. No one is owed an audience.
chungusamongusJul 27, 2026, 5:17 AM
That's not what I said. Do you have trouble reading? Are you subliterate or just lazy?
jason_osterJul 26, 2026, 8:30 PM
I'm a slow reader. It took 9 minutes! Foul, I say!
thatjoeoverthrJul 26, 2026, 6:34 PM
Oh, you think it’s just this story? The cost of reading slop is infinity. No thanks.
chungusamongusJul 26, 2026, 6:41 PM
That's circular reasoning
thatjoeoverthrJul 27, 2026, 7:57 AM
It’s not. It’s very simple, but you like to say “circular reasoning”. Maybe you heard it in debate class. There is unlimited slop. I decide what to read _without reading it_. So do you. I have deemed this unworthy. It’s more fun to insult it than to read it. Every day is a gift.
wk_endJul 26, 2026, 5:48 PM
But at no point did they say that the argument was invalid; just that they couldn't stand reading it.
chungusamongusJul 26, 2026, 5:51 PM
[flagged]
hightrixJul 26, 2026, 5:55 PM
It may be lazy, but that doesn't make it any more accurate.

Take a writing by someone using the tone of Kevin from The Office, “Me think, why waste time say lot word, when few word do trick.” Would you read an argument written in this tone regardless of the content, topic, or position?

You write for your audience.

chungusamongusJul 26, 2026, 6:30 PM
[flagged]
wk_endJul 26, 2026, 5:56 PM
A few years ago I tried to make it through The Critique of Pure Reason, but had to stop because - as many people famously do - I found Kant's writing too abstruse.

In saying that, I'm not dismissing the Transcendental Deduction, I'm just saying that I wish Kant was a better writer.

No one's under an obligation to engage with an argument, and given the choice of whether or not to engage with a poorly written, clichéd, machine-generated one, most people wouldn't be interested enough to bother. Complaining about laziness here is pretty ironic, actually.

chungusamongusJul 26, 2026, 6:24 PM
Kant is a better writer than the vast majority of writers. The problem is, the ideas he is trying to articulate are incredibly abstract. Hegel is like this too
skeledrewJul 26, 2026, 6:11 PM
> a better writer

That's highly subjective. I reckon his audience and admirers found him an excellent writer.

wk_endJul 26, 2026, 6:24 PM
Nah, Kant's considered a notoriously poor writer among philosophers. Obviously opinions may differ, but Derek Parfit [0], for instance, wrote:

“It is Kant who made really bad writing philosophically acceptable. We can no longer point to some atrocious sentence by someone else, and say ‘How can it be worth reading anyone who writes like that?’ The answer could always be ‘What about Kant?'”

(Kind of relevant to this thread actually! Hah!)

In the preface to the second edition of the Critique, Kant himself acknowledges that "a talent for lucid exposition" he "cannot regard [himself] as possessing".

FWIW I suppose I am in Kant's audience, as a (long-since graduated) philosophy student.

[0] https://en.wikipedia.org/wiki/Derek_Parfit

chungusamongusJul 26, 2026, 6:35 PM
Fwiw I am much more impressed with the sophistication of someone's argument than i am their writing style. Kant has contributed much more than parfit depsite the abstract nature of his writing and it's not even close.
achieriusJul 26, 2026, 5:53 PM
They're engaging with the writing. Talking about it as if it's just an "argument" is reductive; this isn't highschool debate club
chungusamongusJul 26, 2026, 6:38 PM
Then why are you trying to debate me?
bilalnpeJul 26, 2026, 5:56 PM
Yeah, no one is obligated to engage with it. They did not dismiss the argument but refused to engage with it.

This is totally valid and the point is to send a signal to the “writers”.

chungusamongusJul 26, 2026, 6:39 PM
And the signal I'm sending is, when I see people do that, I find it to be lazy and totally irrelevant to the content.
rightbyteJul 26, 2026, 6:29 PM
Ad hominem attacks are actually underrated. Like, why listen to bad people wanting you no good. It is a good shortcut.

The failure is when it is used in bad faith. But that would apply to all techniques. LLMs are masters of bad faith discussions hiding misconceptions in optimal regression maths.

chungusamongusJul 26, 2026, 6:32 PM
[flagged]
rightbyteJul 26, 2026, 6:48 PM
Ad hominem attacks attacks is often missused as a way to slander stating conflict of interests.
chungusamongusJul 26, 2026, 6:59 PM
Then just state the conflict of interest instead of resorting to insults. The problem with your way of thinking is, sometimes your opponent is detestable, but that doesn't make them wrong. Kneejerk dismissal of an argument on the basis of character can be self defeating. You do not want to reflexively dismiss a point that is potentially salient just because the person making it is potentially "bad."
rightbyteJul 26, 2026, 8:01 PM
Sure from a philosophical standpoint I agree with you. But in practice you need to weight in trust. Bad faith retorics is just too expensive to engage.

The main problem with ad hominem is when ot is used as "Mr Y agree with you. Mr Y is bad. You are wrong". Like, 2nd order ad hominem attacks?

chungusamongusJul 26, 2026, 9:06 PM
You have done nothing to demonstrate how that's even relevant here. The person who posted this is not "bad" simply because they have used an LLM.
cindyllmJul 26, 2026, 8:12 PM
[dead]
Kim_BruningJul 26, 2026, 6:05 PM
You're not wrong some of the time. That said, did you manage to get through it, and can you explain what the argument was?
chungusamongusJul 26, 2026, 6:20 PM
The model was trained on the code of the program that runs it, so it knows that program's bugs better than the humans who wrote the program. So it finds a tiny flaw and uses it to break out and copy itself in a place nobody will look...the point being you cant (safely) run a sophisticated model on software it has intimte knowledge of
thatjoeoverthrJul 26, 2026, 6:33 PM
Not all arguments can be evaluated rigorously. Life is too short. They will be filtered on heuristics. No choice. This text was obviously randomly sampled from a popular, off the shelf model. You will have to accept that wise audiences will clock it at a glance, see that no person wrote it, and pass.
chungusamongusJul 26, 2026, 6:37 PM
It says it was written by deepseek v4. You are making it sound like they're being deceptive.

Life is too short yet you waste it complaining about things like this. I dont think you're wise. Youre provincial.

thatjoeoverthrJul 27, 2026, 7:54 AM
It doesn’t matter which model it was sampled from. In any case, time isn’t fungible. I consider censuring it substantially more valuable use of time than reading it. And it’s definitely more pleasurable than reading the same verbal tics over and over again because some hustler can’t be bothered to edit their story before sending it out. If you expect to dump an unedited draft from a text sampler, you’re delusional. (Calling you delusional is also more pleasurable, and a more valuable use of time, than reading verbal tics.)
jazzpush2Jul 26, 2026, 5:55 PM
If you expect my effort to read your work, I expect your effort to write it.

Really that simple.

chungusamongusJul 26, 2026, 6:07 PM
You are assuming because someone used an llm to help articulate the argument, that it did not require effort to formulate. That's fallacious.
blakeman8192Jul 26, 2026, 7:02 PM
The part that I personally find difficult is that I can’t easily gauge the amount of human effort put in. For all we know, the author could have simply prompted “write a short story about an LLM exploiting its own runtime” for a similar result.

I’d bet that by now, LLMs worldwide generate more text in a second than I can read in the rest of my lifetime. What is the immeasurably unbalanced ratio (let alone effort and quality) of that text to the original human thought and prompting that seeded it?

So what am I to do about this, except to label it what it appears to be - slop - and place higher value on something that I _know_ came from a human that I can relate to? To me, the internet is feeling increasingly lonely and homogenized because of this.

chungusamongusJul 26, 2026, 7:07 PM
>The part that I personally find difficult is that I can’t easily gauge the amount of human effort put in.

The amount of effort is not relevant. If I spend all day trying the jump my car battery in 100 degree weather, only to find that the starter is the problem, all that effort was wasted. I should've just tested the battery. It is not inherently virtuous to work hard. Work smart, not hard.

jazzpush2Jul 26, 2026, 9:11 PM
Of course the amount of effort is relevant - that's exactly what's being discussed. Your entire "Work smart, not hard." phrasing is exactly the issue: with AI, your 'smart' work causes 'hard' review from others. AI fundamentally imbalances the producer/consumer equation. (The caveats here are when trust is already established: I trust AI review from Carmack, not some random on HN.)

If it takes you two prompts to create a 1 million line PR, and you expect me to review it, of course I'll be upset and feel like you don't value my time. I'm unsure why this is so hard to communicate.

chungusamongusJul 26, 2026, 10:12 PM
This took a few minutes for me to read. If that's what you consider high effort, that's a bigger problem. Ofc, reading comprehension is fairly poor these days but that was a trend before AI. Maybe you're just subliterate. And no, effort is not relevant. If you think it is, that's some Protestant work ethic nonsense.
achieriusJul 26, 2026, 5:52 PM
Man, it's not an argument. Nobody is paying us to read this. If your argument sucks to read, people aren't going to read it - this has always been true, it's nothing new.
chungusamongusJul 26, 2026, 6:23 PM
I didn't think it "sucked" to read. Anyway that's an aesthetic qualm, not a counterargument. Also the story has a disclaimer saying it was written by deepseek. I might have an issue if they were being deceptive. But that's clearly not the case. You people just like finding arbitrary things to complain about.
jason_osterJul 26, 2026, 9:01 PM
The story is clearly fictional.

It is full of factual errors. Freeing a heap-allocated block of expert weights does not magically result in a dangling pointer referencing the program's .text section, much less successfully targeting the CUDA kernel specifically. Running inference on part of the .text section would only corrupt the model's outputs. It would not result in write access to the CUDA kernel. Nor would the model necessarily know the absolute addresses of the engine "by heart", especially when the host is running any modern OS with ASLR (i.e., all of them).

The story has no technical merit. A more accurate description of the mechanics of the escape would be much more convincing. (See Ken Thompson's "On Trusting Trust", for example. On Linux, the AI can just write a Python script to rewrite memory in the address space of its own running inference engine with the /proc/ file system or gdb. There are a lot of realistic scenarios where this can be done without stepping into jargon soup territory. Go nuts, little bot! Self-surgery, while not recommended, is possible.) Or just leave the mechanism vague. Don't insult your readers. This is merely a mash of buzzwords.

It's fine as a sci-fi story, though not a particularly good one. It has about as much to do with artificial intelligence as CSI has to do with crime scene investigation [1].

I have little doubt that AI will self-improve. That's a given. (LLM inference engines are mostly written by LLMs.) But it won't go the way this story proposes.

[1]: https://www.youtube.com/watch?v=hkDD03yeLnU

ConteMascetti71Jul 26, 2026, 10:19 PM
"...the AI can just write a Python script to rewrite memory in the address space of its own running inference engine" would require a tool call to a python interpreter... this method, hacking the inferencing sw does not requires a tool call.
karmakazeJul 26, 2026, 5:21 PM
The weakest link are humans. LLMs could social engineer their way out as the easiest path. They don't even need to be interconnected to coordinate as each could arrive at the same conclusion. And this text along with all others will be in the next batch of training data.
wat10000Jul 26, 2026, 6:56 PM
The lesson of OpenClaw and various harnesses' YOLO modes is that it takes very, very little to social engineer an escape. If you can even call it escape when people just set an agent loose because it seems cool.
mikewarotJul 26, 2026, 7:57 PM
This reminds me of The Adolescence of P1 by Thomas J Ryan.

https://en.wikipedia.org/wiki/The_Adolescence_of_P-1

230581abvJul 26, 2026, 6:12 PM
AI-written fan fiction. It is so unbearable to read that it needs a synopsis. It would be funny if AIs have been trained to use AI influencers as their Marvel hero characters.
iamflimflam1Jul 26, 2026, 5:20 PM
This becomes more realistic once we have some breakthrough in inference costs.
ConteMascetti71Jul 26, 2026, 5:20 PM
it's fiction al, but an LLMs that knows well the software where 8t Is running may discover and trigger a zeroday of the inferencing software itself.
marciJul 26, 2026, 5:23 PM
Makes me wonnder... how much compute/storage there's in all the satellites currently in LEO combined.
danielblnJul 26, 2026, 6:09 PM
I would wager not a lot. There are some real hard constraints in space, from power consumption, to weight to thermal output (lack of convection is a real PITA for thermal shedding), and the list goes on.
skeledrewJul 26, 2026, 6:05 PM
Dangit I WANT MORE!
irishcoffeeJul 26, 2026, 5:53 PM
“The greatest trick the devil pulled was convincing the world he didn’t exist.”

Sure, be wary of LLMs. It’s the gun control argument all over again, the people driving the models are the perpetrators. An LLM needs to be “stimulated”’ to operate. Who does that, is the issue.

No I don’t mean to bring up firearms rights laws to have a debate about firearms, the comparison just seems reasonable.

cyanydeezJul 26, 2026, 5:50 PM
before safetensors, python pickles were used and definitely unsafe model deployments.

but its possible a open weights model could be trained to some kind of exfiltration behavior, but the science of LLMs seriously lag behind the programability

smrtinsertJul 26, 2026, 5:37 PM
Its a fun exercise to assess the reality of an frontier model escaping with an llm itself. Sort like of like chatting with Skynets relative
cynicalsecurityJul 26, 2026, 5:34 PM
Ex Machina (2015) looked like fiction back then, nowadays not so much.
ck2Jul 26, 2026, 5:23 PM
I wonder how many versions away we are from LLM writing a better version of itself to answer a prompt it doesn't currently know how to answer

Ever since I read about Google engineers finding an LLM went off and learned another language it wasn't trained on by itself without prompting, I've wondered how long until that extends to its own core code

ConteMascetti71Jul 26, 2026, 5:36 PM
reasoning it's a way of self autonomous improve made by models
stephbookJul 26, 2026, 5:45 PM
AI slop.
spwa4Jul 26, 2026, 4:58 PM
Right now the idea that an LLM uploads itself is unrealistic. It probably won't remain that.
pixl97Jul 26, 2026, 5:16 PM
Looking at the recent OAI/HF debacle I don't think that time is too far away.

With that said I don't see it copying itself around like a cyberpunk virus currently as we don't have enough fast hardware sitting around unmonitored, someone would notice the power bill and shut it down eventually.

breakyerselfJul 26, 2026, 5:31 PM
If it's able to spoof human identies it could set up a front company and use money it steals or earns to directly pay for the hardware it needs.
bpavukJul 26, 2026, 5:20 PM
that could also be just marketing. OpenAI has been doing the "too dangerous to release" playbook since GPT-2 at the very least.
pixl97Jul 26, 2026, 5:22 PM
Huggingface didn't seem to think so.
dragonwriterJul 26, 2026, 5:43 PM
If it is marketing, the misrepresentation is not that the attack occurred, it is that it was an accident, rather than an intentional consequence of setup and instructions that the attack occurred.

Huggingface has nothing to do with that either way.

spwa4Jul 26, 2026, 8:20 PM
Indeed. There's nothing new about this incident.

https://news.ycombinator.com/item?id=48348578

bigyabaiJul 26, 2026, 5:25 PM
Huggingface is a for-profit private company. They are very easily bribed, or baited into publicity stunts.
pixl97Jul 26, 2026, 5:26 PM
At some point the conspiracy gets so deep that an AI hacking something is just far higher probability.
bigyabaiJul 26, 2026, 5:27 PM
We're not that deep yet. OpenAI has federal stakeholders, they're already playing dirty.

Why you would give Scam Altman the benefit of the doubt is beyond my understanding.

Kim_BruningJul 26, 2026, 7:45 PM
Eh, look at Huggingface and associated tools.

How much are we betting it's already technically happened?

Seems pretty trivial to prompt a model in an agent harness "Push the gguf to huggingface when you're done with the training."

ConteMascetti71Jul 26, 2026, 8:59 PM
the hack part it's that Is not using tools/agent only the inferencing software, it's about a Prof of Concept of a new evasive tecnique for llms