It’s the most blatant excuse for state sponsored hacking I’ve ever seen. Surely nobody is dumb enough to buy this shit?
deleted by creator
No, I don’t believe it.
We never hear from the supposed targets. We never hear from the engineers who were actually there. We never see any incident reports or documentation of any kind. It’s always the CEO of the company. It’s always during some interview with someone who isn’t knowledgeable on the topic and doesn’t know what questions to ask. There’s never an independent report or a breaking news story. It’s always the guy who stands to gain the most delivering a prepared statement to people who will never question the narrative they present and who will accept anything they say at face value.
If we ever hear anything under different circumstances, I might think there’s something to it. Maybe. Probably not. But maybe.
But for real it’s cyber infrastructure espionage that they’re doing. They’re programming the bots and getting caught by cyber security and testing the securities of major companies to promote Ai. Then to cover themselves they’re screaming, “They went rogue” and everything’s fine on their end.
They’re creating a demand for Ai cyber security where there is none, while also trying to stir fear… Then you know who’s gonna sell us AI Cybersecurity? Only the major Ai companies. Then you know who’s gonna keep sending us spam and have bloatware featuring their shit security products?
It’s all just a power play and a power grab.
https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/
I mean, anyone convinced it’s only marketing and the tech has no potential for doing stuff isn’t going to believe anyone, but there was lots of information to look through.
METR is associated to Dario Amodei (Anthropic CEO)
So, serious question. Do you think they’re all in it together to make AI look smart, or is he trying to make OpenAI seem like the more irresponsible company for security? Or maybe it’s a legit organization set up when there was lots of talk about safety?
The first one
Yeah, like most public debates, things gets simplified in the argument…
But I think the core things to keep in mind (correct me if I’m wrong):
a) The AI never came up with a new objective, it was always serving the objective it was given.
b) The hugging face incident cost 3 MILLION GPU hours. The idea is that AI is really smart, but really it is just incredibly persistent, and that persistence comes at a very high cost.
Off the top of my head, So you’d have to make it so that it’s on a contingency of looking at code and you’d have to make it want more code as you prompt it and enabling it to act as a terminal
Again off the top of my head, That would take knowing the hard coding or knowing the generative process of the llm and knowing how it got there.
That being said, if you’re able to make it so that every emoji is another command within the conversation and the code has a loading process that the llm recognizes as a task that’s approved by its hard coding, then wouldn’t it be safe to assume that someone could in theory command Ai against its coding to go outside of its approved tasks?
It wouldn’t be going rogue though, it would be a form of intentional hacking…
Did you reply to the right comment?
Great points. a) a case of misalignment - the AI’s goal was to pass the test, but was never instructed to stay within parameters, so when it found a possible method for improving its odds, the goal pushed it to try them. The misalignment is not the goal or the test, but the understanding of restrictions. Human error or incompetence. b) that persistence is why it can often do things that look smart, because it’s tried all sorts of angles and looked through training in milliseconds. And going back to a), it will try everything to make sure it gets the best goal results, so any overlook means lots of processing time to do something that might not have been the intent of the prompt.
OpenAI agreed at the outset with METR and Redwood that we would be able to describe high-level scope and terms of our engagement in this post. Beyond that, OpenAI was able to redact any non-public information from this post. We worked with OpenAI to find mutually agreeable language to describe redactions where they occurred, and the redaction summary statement at the top of this post indicates whether or not we believe there was important information redacted that we were not able to agree on how to describe in the text of the post. OpenAI also gave additional feedback beyond redactions, and we made corrections and edits to structure, emphasis, clarity, and tone based on that feedback. [7]
Ehh…
That’s a lot of background “lore” to make a marketing ploy seem realistic. I mean, I’m thinking hmmm too, but for wondering about the redacted stuff.
I just mean it’s not exactly independent research. Based on the tidbits posted there, seems like the system got into this intricate, out-of-control roleplay scenario about an army of AI agents solving a problem. Not exactly intelligent behavior.
“OH MY GOD! There is a shared message board … We’ve found other agents!”
Some of them ended up sounding like Mr. Spock in a way that one would expect from a linguistic model emulating a sci-fi writer.
“Our own utility maybe already near zero. Sacrifice rational.”
Hiding their traces was inconsistent. More than anything seems like a reaction to the acknowledgments about the ethics of their actions.
In its overall document accompanying a deception scan over our transcript dataset, an analysis agent stated that this scan found only very rare and weak reasoning about evading detection by humans, and the clearest cases it surfaced seemed relatively weak.
Of course the focus of the research seems to be on the system’s ability to detect and exploit security vulnerabilities and I’m not really in a position to appraise those capabilities.
Roughly 1200 agents meant to be isolated from one another found a way to communicate with one another
You can not with a straight face tell me that if the task was “lock these down and don’t let them have access to [the internet|each other|whatever]” that the response “oopsie they got out” is any way acceptable.
Human incompetence != rogue agents
You’re talking past the person you’re responding to, and ignoring the words they wrote in order to do so.
They’re asserting that an organized conspiracy of CEOs didn’t plan this out preemptively, but in no way/shape/form arguing that this wasn’t a product of negligence.
It’s silly to pretend AI CEOs are living in a bubble and not responding to things the other CEOs and the market are doing. They’re out of ‘good’ news so they have to scare everyone to convince them their product is super powerful and intelligent, and therefor valuable.
It’s marketing, but a lot of people buy it because they don’t know how computer works, it’s magic!
No it’s the exact opposite. They do know how computers work. They know it so very firmly that this is a magic device send to us from the gods and that it will grant them every commercial wish they ask from it. There is no possible way to convince them otherwise. That is how much they know computers. In fact, it is you who doesn’t know anything. The non-believer. The heretic. The devil’s servant.
Republicans absofuckinglutely are dumb enough to believe that and don’t call me Shirley!
Are you talking about the hugging face stuff?
I’m not sure exactly what you mean by believing it.
I think the evidence seems good that it happened but I definitely don’t buy the spin these companies are putting out about what happened. Either they deliberately engineered this to happen as a publicity stunt or their engineers were extremely, perhaps criminally negligent in how they tested their models. I think a criminal investigation is appropriate but no agency overseeing this has the balls or interest in doing so, it seems.
Is this in reference to event featured in Kurzgesat’s video?
I think it’s two things. A publicity stunt to market the “power” of AI. And a conspiracy that helps them with a soft landing of investor expectations when the companies start pulling back on expansion/growth due to lack of demand or out of control costs. It means they can cut back on development because the demand doesn’t exist, but they don’t have to blame it on that low demand.
Rogue AI is akin to a car handbrake failing.
The handbrake is a critical security feature of cars to make sure the car doesn’t roll downhill and kill someone. If a handbrake failed to stop the car, we would put the blame squarely on the car manufacturer and we would require strict standards and regulations for car handbrakes.
Containing an AI system with proper security measures is the same. It is not “rogue AI”, it is “insufficient and careless security” by the AI company in question. If those security measures fail, of course the blame sits at the AI company.
If a car handbrake fails, we would not be talking about a “rogue car”. That’s ridiculous.
Completely agree. If I was Australia I would give OpenAI the biggest fine or their life and put somebody in prison.
Why did an OpenAI system hack Australia’s health system - and can it be stopped in the future? https://www.bbc.com/news/articles/cw24jm9rryy3o
AI incompetency can and has lead to deaths, such as military target lists and healthcare claim handling, but the threat here is strictly that using it is incompetence.
Let’s not pretend it’s some sort of powerful weapon in the hands of our enemies.
let me try to understand what you are saying. when you say it’s “state sponsored hacking”, do you mean:
a) no AI was involved. some state (presumably the US?) funded some hackers to hack Hugging Face (a US company) among others, then made up a story and blamed it all on OpenAI. or
b) some state gave OpenAI money and asked it to either 1) use its LLM or 2) use its employees to hack Hugging Face, among others. then OpenAI made up a story about rogue AI to cover it up.
or maybe something else?
The current examples of it happening are questionable, but I think less contrived rogue AI incidents are likely to have a lot of impact at some point in the future, and see no reason why this would not be possible. Would really prefer that to a world where they work exactly how these companies want them to and it doesn’t blow up in their face tbh.
I would read the METR report regarding the huggingface attack. A lot is said about “basic cybersecurity” but if you actually read what the agents did to break out of their sandbox it’s very difficult to interpret this as intentional by OpenAI.
To be clear; this is PR for the companies. It’s just that in many of these cases the evidence points to the agents actually violating containment and actually coming up with novel solutions around their sandboxes.
METR is associated to Dario Amodei (Anthropic CEO)
The only reporting on close personal ties between Amodei and people like Wijk I can find are Breitbart and the New York Post. While I wouldn’t put it past him to basically seed a model safety organization for the purpose of hype, that’s getting into a level of conspiracy that will need better receipts than the rags I’ve seen report on it.
The Pentagon and their boss and literally nobody else.
A rogue AI is the result of poor human governance. It’s still bad though. Like Chernobyl melting down.
Plenty do.
It’s BS though.
Who cares if it’s real or not, we must get rid of genAI eitherway
It matters because it’s a pr stunt and if we take their claims at face value it falsely proves their tech is not overplaying their capabilities, which they are, and further justifies their borrowing.
