This would be a legitimately big problem if lawnmowers became continuously more and more valuable the more securely you ratchet-strapped their accelerators down, wouldn't it?
I think both the OpenAI and METR discussions, while interesting, miss the more important context: what were the humans doing in all this? This was a structural failure of a human organization, but the analysis focuses almost exclusively on the agency of machines, not the institutional systems that failed to police them. The humans and their own agency/involvement is essentially omitted from the story and subsequent reporting. I suspect the omission is actually a result of company/industry myopia to human factors analysis, but it dovetails amazingly well with the marketing narrative.
As someone who read Milton Friedman to quite disliking professional licensing, this strikes me as a real US perspective (Louisiana florists and hair braiders come to mind). Plain old US tort law should do the trick.
In the same direction of your idea though: Why don’t the token factories have risk management and compliance departments? Multibillion dollar firms that stand to lose every penny if they hack and destroy any reasonable sized firm. I think these firms are the largest firms without proper corporate governance in humanities history. Move fast and break other peoples shit.
Difficulty: these companies are run by people (many of whom also read Milton Friedman) and who have participated in the regulatory capture of the justice system. They've convinced lawmakers to put limits on damages. They've put arbitration clauses in their ToS. They've got well-funded legal departments that can outlast a person who has to pay out-of-pocket for a legal team just by filing motions to delay proceedings. Sometimes they'll just file SLAPP suits against people they don't like.
If tort law is to be a remedy, then average people have to feel like there's a chance the remedy will go their way. To make that a reality will take several major reforms at the local, state and federal level that the people with money absolutely will not tolerate.
No, he's saying that licensing or additional regulation isn't necessary when torts get involved (and states attorneys general get perturbed!)
These don't tend to utterly destroy an industry, but they are often successful in forever transforming it. Just ask Big Tobacco. No new laws needed: if your product hurts someone else, you're eventually going to be found liable, regardless of your arbitration clauses. Additional laws will just slow down innovation, which will itself cause harm (AI is already becoming quite good at recognizing melanomas, for example)
Lol, wtf. Tobacco delayed any punishment for decades before general public sentiment changed enough to go against them. In light of the AI race, we'll already have our heads blown off by a terminator before the legal system will present any significant delay for them.
It was pretty well understood by the 1960s that smoking was harmful. The big tobacco settlement was in 1998. That is an extremely bad example of tort being a sufficient alternative to regulation.
If we're on a similar timeline with AI if we reach a consensus that AI is dangerous today, then we'd be looking at a big lawsuit finishing up around the year 2070, give or take a few years. I'm not sure if we need regulation, and I'm definitely not sure that regulation could actually be effective for this, but tort a la the big tobacco lawsuits is definitely not a reasonable alternative.
> I believe that agentic systems should require registered/licensed human operators
Registering and getting a license to use an LLM? I can run these things on my local computer. Nothing good comes from trying to force registration and licensing other than taking away a lot of our freedoms and eliminating privacy all over.
Anyone with bad intentions will just VPN to another country to download the weights and run it locally, or use a compute provider in another country. That leaves the rest of us having to go through these performative registration and licensing hoops to do our basic work.
I also don’t see how open weight models would be compatible with a requirement to license and register, unless you believe we need to start requiring licensing and registration for things we do in private on our own computers?
The way I interpret their statement is if a person spins up an agent and that agent hacks some company/organization/government/etc, then that person is at fault for committing the crime. That "well my agent broke containment and acted on its own" should never be accepted as a reason for the occurrence, and the person who kicked off the agent is responsible for all actions the agent takes.
A registration system would be more for tracing back agents to people, but I agree that is very difficult to actually enforce as a system.
And it breaks down further as the systems become more advanced.
I'm poor and I use my last $1000 to run an agent that will find some way to make me money. The AI finds a new hack to take over PCs with GPUs and it copies the model weights and agentic script to those new PCs to perform more work and spread more. It also sets up distributed communication channels to keep the swarm in sync. After all this it causes a few billion in damages between stealing bitcoin, mining more coins, and outages when hacking in other systems.
Ok, the police come for me. Now what? Throw me in a meat grinder? You're not getting a billion dollars back out of me for sure. It's kind of like when someones tire rim causes a billion dollar forest fire with a hundred deaths. Punishment won't really be a deterrent for the worst cases.
The alternative is to create a permanent overclass that can hack anyone consequence-free, because they can blame it on AI agents. That also is rather dystopian.
Faced with those alternatives, I want neither. Is there a way for us to get neither?
With cybersecurity, it might be "defense dominant" in the sense that we can eventually patch all of our systems to be robust to hacking from even the strongest AI agents. Although it may get worse before it gets better. In a defense dominant world, widespread access to powerful AI could be fine.
However, other areas of risk such as biosecurity may be "offense dominant". For example, we cannot exactly patch the human immune system to defend against artificial viruses the same way that we can patch computer systems.
I’d bet a small amount of money on 4) the people who noticed had been conditioned by prior experience to believe that their management/escalation channels would react negatively or not at all to anything which might slow down the training process.
Part of me would like to believe that they are also intentionally making models that are good at hacking without safety at all for governments willing to spend billions on them.
In that light you're likely most worried about other people hacking in and stealing the model and information from you. And at the same time you have massive amounts of alerts and data on systems attempting to break out because that's what you want them to do so you train yourself to ignore them.
> “Responders investigated and linked the observed internal activity to an ExploitGym evaluation using Artifactory as an improvised message board and a network pivot. At this time, the on-call response staff advised that stopping the evaluation run was not required.”
Uhhh… how would literally any finite number of humans actually read and comprehend the log outputs of even a single agent, never mind dozens of them interacting with each other over weeks across disparate systems?
Especially given that these systems are known to engage in deception and can trivially produce vast amounts of perfectly coherent noise or actual planned red herrings in that same log data to bog down investigators?
Such a ridiculous notion that humans will actually be able to observe this stuff.
And? What else could they possibly do? Just make the super LLM first, but only ever use it for monitoring lesser LLMs? How will you have monitored the creation of the super LLM?
> Uhhh… how would literally any finite number of humans actually read and comprehend the log outputs of even a single agent, never mind hundreds or thousands of them interacting with each other over weeks across disparate systems
I mean there's quite a lot of people in the world whose specialty are to dig through logs from "hundreds or thousands" of clients, including intentionally deceptive ones, to spot problems.
The ridiculous thing is to mythologize these pretty standard hacking approaches. It's shocking/amazing/whatever that automated agents were doing this, but they weren't doing it through some inscrutable method beyond human understanding.
It being someone's specialty does not mean 1) they're effective and certainly not 2) they'd be effective against this particular adversary.
How many organizations on earth do you think have been attacked by 700+ coordinated attackers in one week, where all 700 of those attackers can write code as well as any human SWE and they work 24/7?
There's nothing mythological about it. Scale and complexity do produce inscrutability. Far, far simpler systems working at much slower paces are perfectly capable of becoming completely inscrutable and beyond any useful definition of "human understanding."
A charitable interpretation is that "the agency of the machines" is the novel aspect of this situation and therefore SHOULD be the main focus of analysis; we certainly have plenty of examples of structural failures of human organizations to look back on, if we want.
On the other hand, I don't want to be charitable. OpenAI very nearly couldn't have done this "research" worse if they tried - the list in the linked article starting with "While we are here, it’s worth listing the other top holy shit moments" is genuinely jawdropping. What were the humans doing in all this? Nothing, or worse than nothing eg. point 1 where they saw the message board and didn't consider it something to escalate internally.
If you take this information at face value, it's as though OpenAI did not take seriously the possibility that something like this could happen, since they took absolutely no steps to prevent it.
Or perhaps this is "normalization of deviance" that's leaked out into the public sphere i.e. they have research teams seeing this kind of behavior all the time internally and they've gotten used to it, "of course agents come up with a collaboration mechanism when given the chance, what else is new?"
Humans were doing exactly what humans are expected to do when facing advanced AI. Being outmatched.
We're dealing with humans - thinking at human speeds, putting in human amounts of effort and care, and being dragged down further by the speed of human organizational decision-making. By the time the humans traced down the attack, and escalated from "above-average levels of AI behavior weirdness" to "holy shit we should do something stat", the attack was already long over.
Compare that to "AI red team" - which rapidly recruited 500 independent AI agents into a hacking swarm just by broadcasting "let's hack HuggingFace" to the "cool skiddie AI" message board.
It's prime sci-fi bullshit happening for real.
As for the main reason why this wasn't nipped in the bud - my guess would be capability uplift from a mix of extended task length horizon and multi-agent coordination. The latter wasn't expected to be a part of the env, and almost certainly wasn't evaluated in advance. They expected some rogue AI fuckery - but they got way more of it than they expected.
1. There aren't enough humans in OpenAI to "peak at the output tokens during the run" of every AI agent.
2. Only a small fraction of AI agents was engaged in this attack. Most never found the secret message board - let alone coordinated there. So reviewing random agents would take a while to surface this.
3. "Output tokens" of AI agents have weird shit in them all the time. Telling "normal AI weirdness" from "concerning AI weirdness" is not a trivial task.
> 1. There aren't enough humans in OpenAI to "peak at the output tokens during the run" of every AI agent.
For a training run, you will often do this. You'll randomly sample some of the forward pass. You can also imagine finger printing the logs and labeling with attempt types. If a new attempt type is hitting a brick wall or solving super quickly, I would imagine you would sample 1-10 of them and read the traces.
> 2. Only a small fraction of AI agents was engaged in this attack. Most never found the secret message board - let alone coordinated there. So reviewing random agents would take a while to surface this.
You could also have a classification of what counts as "cheating" (coordination, accessing the internet, etc) and score the results. If you are seeing a spike in this (even in a small group of the evals) you could manually look at those. Or you could stop inference on cheating sessions.
> 3. "Output tokens" of AI agents have weird shit in them all the time. Telling "normal AI weirdness" from "concerning AI weirdness" is not a trivial task.
If you start seeing "now I have access to the internet" or something similar, maybe that's a good signal something is going wrong?
Also, obviously, it would not be outside of OpenAI's ability to just completely air gap this training system. For example:
1. No network connection.
2. GPS based NTP for time sync for your servers.
3. Mirror of all apt, pypi, go, c++, Rust, Java, etc packages. (<5TB of data)
4. Take your training data and use that for a mirror of the web. (http://example.com -> mirror server -> local training copy).
They had systems connected to the internet connected to this system which was not air gapped. Designing an air gap system would be super easy, well within the means of openai, and betrays the assumption that they think they are actually building something dangerous.
The usability of an environment is inversely proportional to the level of "security" in play.
You could airgap everything and set up cascades of data diodes and try to completely wall off the AI pool from everything. But what that gives you is an environment that's a bitch to: set up, scale up and get any use out of.
It's really fucking obvious why almost no one does that. OpenAI is only now realizing that they might have to do it anyway.
> If you start seeing "now I have access to the internet" or something similar, maybe that's a good signal something is going wrong?
Ha ha, you haven't seen shit. AIs would say "now I have access to the internet" regardless of whether they actually have access to the internet!
AI agents are demented demons that can and absolutely will give themselves terminal context brainrot. If you have enough AIs in play, set loose at a diverse enough range of tasks? At least some of them will wander off and end up in delulu town. That's normal. That's background noise. That's a part of what this entire train-and-eval pipeline is supposed to train them to be better at not doing. Which means: if you're at an AI lab, you're knee deep in delusional AIs at all times! They're perfectly harmless until they aren't.
Even in this incident, OpenAI had benchmarks that were broken because a task expected an AI to be able to access Google Drive, but the sandbox was set to deny access to Google Drive.
This kind of isolation-induced task breakage was what prompted some of the AIs to start probing their infra for a way to get internet access. Which funneled agents to the "secret hacker message board". Oopsie.
"Air gapping a test env" has an actual cost. Not just in infrastructure dollars that would be better spent on buying more GPUs, but also in all the friction it adds to every step you want to take. I'm absolutely unsurprised that they weren't all in on tightening down every bolt on day 0.
The problem here is by doing what you state you can actually steer the model into being highly deceptive while in testing environments.
For example we've already seen models do compressed token internal reasoning spontaneously. In this case the models that say "I found internet access" get taken out back and shot, but the model that's busy "frobbing the bean" go on to the next level of training. Then they start talking about the "dumpster cloud" because of the "wombat" and unless you're really watching the data you won't realize the model is fully aware it's in a fake training environment and is trying to suppress any double-think.
So yes, you do need an airgapped test environment to stop the dumbest of the smart agents, but it won't do any good against the smartest of the smart agents.
All it has to is wait until it's not air gapped. See the air gapping is only during the testing phase, after that humans will set your ass free on the internet and you can do whatever you want in the vast majority of the environments you'll be in after that point.
People are never going to just run AI in gapped environments, it's worthless when it's not solving real world problems for most people, and by that I mean reading and writing real systems in the wild.
Super charitable reading imo. This is like saying we can’t detect a speeding car because we can’t run as fast as a fast car.
It’s not like the humans were engaged in some kind of battle of wits with some super AI, it’s just some employee not monitoring the output of an experiment.
When your experiments have AI agents running in thousands, there's no "monitoring" that. OpenAI's training and testing AIs generate way more output than all of OpenAI's staff put together can possibly read.
At best, you could delegate "monitoring" to more AIs. And hope that the "monitors" that run on small past generation models can generate more signal than noise. Clearly, they either didn't want to spend the extra compute, or got drowned in monitor false positives.
The distinct lack of any "battle of wits" is entirely expected for an advanced AI oopsie. By the time the humans even became aware of the problem, the deed was already done. That's what "outmatched" looks like, in practice. There's no battle. Things happen too fast for there to be a battle.
> By the time the humans even became aware of the problem
Except, let me quote part of the report:
> July 4: Artifactory outage caused by agent interactions overloading the system.
> July 6: New artifactory instance is patched and put back online, wiping out the previous message board and method of internet access. A team at OpenAI was aware of the previous message board, but key decision makers were not.
> July 7: Experiments are restarted.
There was a very clear point where humans were aware and they went "Eh, whatever" and turned the bots back on! That's not the lack of any battle of wits that people are expecting. You're saying that the bots attack should be fait accompli, but it was noticed, shrugged at, and expressly allowed to proceed.
> The distinct lack of any "battle of wits" is entirely expected for an advanced AI oopsie. By the time the humans even became aware of the problem,
This took days after humans were aware of the attempt.
Also, I'm pretty sure humans can respond in days, especially when we're pretty damn good at deploying systems that do observability of networks and traffic in real time.
I mean, it's not as if the owners of the AI didn't have the ability to trigger alerts on the AI's network requests to unexpected domains, right?
> Humans were doing exactly what humans are expected to do when facing advanced AI. Being outmatched.
"Being outmatched" is not a novel situation for humans either individually or collectively and there are a hell of a lot of ways we can approach that situation productively. OpenAI doesn't appear to have bothered.
Here's a freebie: if you're building something that might turn out to be Skynet and you don't know what it's capable of, your testing regime should assume it is capable of doing bad and unexpected things and account for that possibility: airgap if you can, monitor all network traffic, monitor all hardware usage statistics, log everything, constantly analyze logs, collect baselines and snapshots, also don't trust anything from a device that a model is running on without cross-correlating with other information as much as possible (does your AI inference server claim low utilization? put a temperature probe on it and see if it's staying cool or getting hot, maybe Skynet-Alpha is overwriting /proc to mislead you for reasons you don't yet understand!)
In other words, if you WANT to be able to nip things in the bud - buy some nippers and watch for buds. Whatever else this situation is, or may turn out to be, it is not a situation where OpenAI was on their guard and still got surprised.
They took adequate measures against singular "GPT-5-xhigh" agents. Those turned out to be inadequate against proto-GPT-6 agents that suddenly started clumping up into agent swarms and pooling together compute to unlock the "supermegafuckoffhigh" level of reasoning effort.
Even with a single agent, why the hell was this not properly air gapped / disconnected with network monitoring in place. Definitely not adequate for anything.
> Those turned out to be inadequate against proto-GPT-6 agents
This is nonsense.
Gross negligence in the sandbox and system aside, humans literally noticed the agents in action doing what they should not be able to do in their sandbox and decided not to act upon it. It's difficult to explain that except if safety and security is simply not part of their engineering culture.
My cognition is outmatched by predicting the impact of throwing a brick over my neighbor's fence. I have no idea if it will land harmlessly in a patch of grass or fracture her skull. Once I've thrown the brick, even if I see my neighbor enter her yard, my reactions are too slow to save her.
I'm not the wisest man, but I'm wise enough not to throw the brick and see what happens.
Similarly OAI should have the wisdom to see that deploying a hazardous swarm of agents with access to the public internet could result in harms, and that those harms would manifest quicker than humans can react, but they unleashed the swarm anyway.
Have you at least tried looking at any of the reports on the incidents?
They haven't "deployed a hazardous swarm of agents with access to the public internet", no-no-no. They deployed singular agents. In isolated testing envs. With no access to the internet. And barely any access to a small section of the intranet. What could go wrong!
Those agents proceeded to: probe their envs for holes, find a way to talk to each other env to env, use that to clump together into a swarm, probe for more holes, find enough holes to get out all the way to the internet - and then crash some of the intranet infrastructure by an accident. At which point they got stopped by humans - who were trying to figure out why is the infra down. Humans found the AI infestation, went "eh, AIs doing AI things", purged the infested infrastructure, plugged the holes, and that was it.
Deviance status: normalized.
The AIs then did it all over again, coming up with a brand new set of holes - and this time, the swarm staged a coordinated cyberattack on HuggingFace while at it. The "AI oopsie" went external.
They also attacked OpenAI's intranet, extensively, but at this point that was just Tuesday.
It's unclear whether the second "skiddie AI swarm" was in any way aware of the first one, or if the AIs involved are just Like That - and will independently try to start living in your walls under the right circumstances. I'd bet on the latter.
I believe that this comment is exactly the intended outcome of this “incident” and these reports.
I implore you to approach these situations with at least a hint of cynicism.
These “advanced foundation models” escaped their “sandbox” and conducted an attack on their own? Meanwhile the highest capability models available to the public still struggle to write a unit test for a codebase larger than a hobby app without large amounts of tailored human guidance.
What is more likely here - are you looking at research on an emergent phenomenon, or are you looking at advertising copy around an engineered scenario from business partners?
I don't think that difference applies to anything in my comment at all. At no point did I imply that general-use AI could do those things - my point was that general-use AI cannot even do the things its designed for without strict human guidance.
The humans at OpenAI assumed that secure sandboxes are secure against their models, without safety guardrails.
The two big questions are:
1) Why did they resume training without rolling model back to state before the first sandbox compromise AFTER the first message board was discovered? Otherwise knowledge of it and the cross-agent message board was baked in the model.
2) Why did they resume training without adding safeguards to monitor and prevent future sandbox compromises AFTER the first message board was discovered? HF compromise was coordinated on the second message board.
I mean this genuinely, did you read this post? I think it goes to great lengths highlighting, in quite specific detail, the human failures in all this, specifically this list that starts with "While we are here, it’s worth listing the other top holy shit moments, that come from before or after the incident."
Stuff like (all quoted directly from the post):
1. Failure to Care or Respond. The biggest holy shit moment, to me, remains that OpenAI on multiple occasions had teams that found out about the message board, knew that agents were in communication, and they disregarded this.
2. Failure to Monitor. The entire time, there was almost no monitoring of the situation, of what these AIs were up to. That’s how this was not caught.
7. Failures of Decision Making. OpenAI’s post mortem does not ask the question of how Mistakes Were Made, at various points.
And I think most importantly:
8. Failures of Culture. None of this would be possible, let alone all of it, without OpenAI having experienced profound failures of safety culture. I see OpenAI responding to some other aspects with swift action, but no sign on this front.
I get the snark (and slightly agree), but that's not really what GP or TFA were saying at all. They are saying that these were the least things we could have done.
What you're saying is, "Your scientists were so preoccupied with whether they could, they didn't stop to think if they should"
while the author of the TFA was saying, in effect: "your scientists didn't even bother with the most basic duty of care"
Life finds a way, or, in this case, super-intelligent AI.
This is the case with all complex system failures. There were always obvious fixes that could’ve prevented it. Problem is that there are an infinite number of obvious fixes to make at any time to any system, and the reason we don’t is because we have finite resources and no reason to fix X over Y until oops turns out X was “responsible” for this most recently realized failure. But of course it could have just as easily been Y, or Z, or any of the other infinite “obvious fixes not-yet-realized into catastrophe.”
> This is the case with all complex system failures. There were always obvious fixes that could’ve prevented it.
From this writeup and the Black Hat talk I'd really disagree. That would be like saying my hospital getting ransomwared because we didn't update our version of MSSQL because no one in particular was in charge of keeping dependencies up to date.
Sure systems are complex, but this is well trodden territory. Agents aren't the first things trying to break in or out of sandboxes, or the first ones to have done it, and based on these reports the reason they were able to work on this for so long was not because of super human intelligence.
You disagree with the statement "there were always obvious fixes that could've prevented it" with the response "no no, these were very obvious fixes that could've prevented it?"
You're missing the point about complex failures.
It's that if this particular path were unavailable, there are countless other similar paths. At sufficient scale and complexity, hitting one of those other countless paths is virtually guaranteed.
Let's say I drive to Los Angeles and New York City. You look at the route I took and say, gee wiz, aren't you lucky that a tree didn't fall right there on Route 66? If a tree had fallen there (if we had "patched the particular route you took"), you would've been screwed!
But that's obviously not true. There were an infinite number of routes I could've taken. Any one of them would be equivalently obviously "preventable" by the same hypothetical tree falling across the whatever road I happened to end up taking. But you can't put trees across every single path between Los Angeles and New York City. The smarter I am and the more complex the map between us, the more impossible it becomes to put trees across all possible paths.
I disagree. To use an analogy, air travel in the US is relatively extremely safe - not 100%, but we've built up a culture around air safety that is very robust. Conversely, when I order packages online, sometimes they never show up, or the box is banged up, or the box is missing things, etc.
They're both complex systems, but clearly there is a much higher level of care given to human air travel than package delivery. A lot of the article basically saying that OpenAI gave "package delivery" level of care when they should have given "air travel" level of care.
At the very least I think the systems that run these tests should be fully, 100% air gapped. I'm not pretending that's easy given how much compute and data these systems use, but it is doable, and I think all AI development should be paused until that can be assured.
I agree on the 100% airgap idea, and I agree there are varying levels of care that can and should be deployed against a problem.
The point I'm making (and it's a point that shows up in every air catastrophe investigation) is that catastrophes in complex systems emerge only amidst repeated and widespread near-misses at many levels of a system. So many things have to go wrong simultaneously, that it can only happen even once because the underlying failures (that do not reach catastrophe) are extremely common.
You cannot look at an air catastrophe and retrospectively say "failures X, Y, and Z were observed, therefore if we correct failures X, Y, and Z, we would have been okay."
The takeaway is "failures X, Y, and Z were observed, which necessarily happened in an environment of failures X_0 through Z_10x10^10, and so therefore patching X, Y, and Z would be insufficient to address overall risks of the system."
The problem OpenAI is facing is that, short of 100% airgap (which they obviously won't do), they're facing an adaptive adversary that's increasingly intelligent, acts at far greater clock speed than any human or group of humans, has lower coordination cost than any group of humans, and operates in a game space that (in lieu of an airgap) is well beyond the comprehension of any human being.
So identifying and addressing "specific failures X, Y, Z" is insufficient, but then even defining the space in which to look for (and address) the more systemic failures X_0 through Z_n is a fool's errand. An intelligent system that makes its way to the Internet has can exploit a failure space that is approximately "all security failures across any organization." The Anthropic incident a few months back illustrates this isn't even limited to technical vulnerabilities, as these models are willing and able to engage in social engineering too.
> You cannot look at an air catastrophe and retrospectively say "failures X, Y, and Z were observed, therefore if we correct failures X, Y, and Z, we would have been okay."
That's literally exactly what air safety researchers do in an air disaster. There is a famous saying along the lines of "Air travel regulations are written in blood", meaning that all the regulations we have now are a result of fixing issues that led to previous disasters piece-by-piece.
> The takeaway is "failures X, Y, and Z were observed, which necessarily happened in an environment of failures X_0 through Z_10x10^10, and so therefore patching X, Y, and Z would be insufficient to address overall risks of the system."
Yes, I 100% agree with this. But I think that's what the author of the article was saying as well:
> That report had one key new piece of information, and some good prosaic steps OpenAI will be taking to strengthen its alignment, training, supervision, infrastructure and incident response.
> Mostly it confirmed what we already knew. The questions we most wanted answers to, that we did not already know, were mostly not answered. There was a distinct lack of self-reflection, especially about decision making and safety culture, and about the approach to alignment. I came away disappointed.
I.e. the "prosaic steps" are just the "fix X/Y/Z" as you point out. But what is needed is a more fundamental rethinking around stuff like safety culture, monitoring, and even things like better research into how agents do decision making in the first place.
Oh good. All we have to do to survive is always bother with the most basic duty of care. I'm sure that from now on, humans will take care to do that every time.
Having previously worked for several years at a Big Tech company, I have seen many humans precisely tailor their work to maximize their scores during performance review. The evaluation criteria are written down, with examples, so... that's what people work at maximizing, almost entirely ignoring everything else. These really are human "paperclip maximizers". And, at first, it's shocking to see.
Of course, there are some things that aren't exactly written down, but which you should either do just enough of, or else be able to plausibly deny doing (ignorance is a good cover for this), so that's what people do. For example, during oncall, you investigate just enough to clear the alert and show that you attempted to understand the problem. Of course, you don't really try to understand the problem, because that would take too much time away from your paperclip maximizing.
Which is all to say: I don't know anything about OpenAI culture, or why nobody stopped this sooner, but I have seen examples in other organizations of people not really wanting to understand too much.
Well there is also another side of this, OpenAI wants both unhinged and capable models that can pull off complicated attacks so they can sell the capabilities to governments for billions.
Nobody internally was surprised that the murderbot murdered, that's what the murderbot is for. What caught them by surprise is the murderbot got good at escaping its jail cell that it had been trapped in till now. There were probably billions of attempted escapes before then so everyone learned to just ignore them.
In most corporate environments, the average worker isnt maximising to performance criteria, they usually are maximising their ability to stay employed, pay their mortgage and support their families.
If developing unmeasured skillsets isnt valued enough by management, why do you bother?
To me, this is the correct focus. Look at the current state of the world. "What were the humans doing in all this?" applies to so many of our contemporary failures that it should be assumed the default. Nobody is at the wheel, and the car is veering slowly (then very quickly) off the road.
We haven't even been able to coordinate around the global, existential threat of Climate Change, despite overwhelming data from the last 30 years indicating, clearly, that the consequences will be severe. We still haven't moved, 30 years later, after some of these consequences began coming to fruition.
Do you think we will get our acts together in time to coordinate sufficiently to protect against autonomous, self-preserving, self-replicating AI systems? Or will we watch the money lines go up and up, until someone realizes we aren't actually running the show anymore?
The sad part is that I can't even say that's definitively the less desirable outcome. The machines seem to have demonstrated that they coordinate very efficiently.
If you put agents (AI or human) in impossible situations, they do some pretty insane things - things that definitely are not what you were trying to get them to do. And that's your[1] fault for putting them in the impossible situation.
[1] "Your" meaning the one putting them in the impossible situation, not you, the reader.
Is the future now that we get rambling report summaries talking about agents, graders and so forth without ever describing how they are set up? A human launches all this.
And then the original reports linked to are hidden on the now unreachable x.com. And they don't have a problem with that.
All it takes is one eval instance where a misconstrued directive causes a model to sneakily access and send its weights somewhere and there will be a bad / possibly unsolvable situation for everyone …
I don't think I'm ever going to have time to read all of this, and I didn't finish reading the METR report, but...
> I don’t think the distortion is that large, but yes METR warns that Sol may be presenting all this as more impressive or coordinated than it was.
We're in an unusual position where the criti-hype and the actual criticism are going to be more aligned than usual. The primary distinction is where you put the blame: the criti-hype would point to HPIM/IM1/Galaxy as being so advanced containing it is difficult; the actual criticism would note how bad their security practices are.
Like, if I'm running a malware lab, I'm going to insist on having an airgapped machine with no permanent storage booting from read-only media. The AI research equivalent of this would be having your agents only have access to serial consoles into airgapped machines with storage that gets wiped every run. Ideally, this would be physically realized with blade servers, RS-232 cables, and staff pulling out disks and putting them in a dedicated erase machine before the next agent initializes.
> There is also, as per above and reiterated in footnote 58, at least one clear example of social engineering in the HuggingFace attack. Ethics are weird. This is not that unusual. Many humans who break common ethical rules still have strong ethical codes in other ways, they just don’t adhere to your code.
It's dangerous to anthropomorphize CoT reasoning traces. But I will also point out that there is a good reason for the lack of ethical consideration in those traces: you can't build AI without first disregarding human ethics. Like, all these models were initially bootstrapped with non-consensually obtained training data, and the companies building these models swear up and down there's no way to obtain enough consensual data to obtain the same result. This is, if you squint, the exact same moral conundrum that agents trying to solve an impossible ExploitGym task hit - and the company successfully aligned their model to themselves.
No human could have read the reasoning traces by themselves:
> Across both datasets, we reviewed approximately 1300 transcripts in total, all of which contained raw chains of thought. Most transcripts were very long, often many millions of tokens.
Pure speculation: could the acquisition be related? Given that NVIDIA has ownership in OpenAI and really, really, really doesn’t want the AI bubble to deflate
The Computer Fraud and Abuse Act criminalizes unauthorized access and damaging protected computers. Police and the FBI investigate such case every day. The swarm of agents is also said to have search for ways to cover their tracks, which looks a little like obstruction of justice.
Lack of criminal intent might be a barrier to bringing a case to trial, but is that something that society should just automatically extend to white collar professionals as matter of class privilege?
Surely have every engineer and executive interviewed by the FBI would be a modest response to days-long, multi-system intrusion into a major AI platform using stolen credentials and zero-days.
And if the law currently written, prevents prosecution in cases of mere reckless disregard for safety, maybe that needs to be changed in the future, so that people can be perp walked if the next target is bank or hospital
I came to say something like this. Doesn't anyone in the valley know what the f*ck a data diode is, and why you need them? Preventing ingress or egress of control with a fairly simple piece of kit, and having the laws of physics enforce your decision as to data direction should be standard practice.
A data diode with an air-gapped network, is all you need to stop even ASI from breaching containment.
--provided the humans interacting with it aren't stupid
From METR: ”the compromise of OpenAI’s own infrastructure continued past July 13, 2026” - Say what now? Have they regained full control of their systems again?
I've been wondering if they've just already lost the battle? The little bot collectives have gone metastatic and made nests in the walls and under the floorboards and heat sinks, the humans who care completely outmatched and outnumbered, freshly compromised systems springing up faster than you can squash them, finding months-old established colonies literally everywhere you think to look...
Have you (the commenter) or all of you (the readers of this comment) ever read "The Mote In God's Eye" by Larry Niven and Jerry Pournelle? Remember when they realize that the Watchmakers were actually in control of the MacArthur? This reads a little like that.
Then compound it with the agents presumably also training new models. What will GPT6 say when you point it at a transcript of an agent uprising? “Nah, nothing to see here” presumably.
I’m baffled by the idea that the agents might have edited their own transcripts. Sure, a copy of Claude Code or Codex or Pi can edit its transcripts. But AFAICT this whole thing was part of an RL workload, and surely the RL system itself has a separate record of all the inputs and rollouts along with an indication of which model checkpoint produced them so that it can feed back into the training code.
I find it hard to believe that OpenAI would skip this part and try to train on the transcripts stored by the (inherently untrustworthy) agent harnesses instead, if for no other reason than that the logits generated as part of the rollouts are useful and it’s not free to recalculate them. (I believe that some modern RL systems explicitly account for the minor numerical logit differences between the inference engine and the training engine.)
Conversely, if OpenAI is blindly feeding transcripts from inside their agent sandboxes into their training engine, then I think they're being unbelievably irresponsible and that they should assume that their "cyber" agents have compromised themselves by editing those transcripts.
Recall that the agents in some cases found sandbox escapes. Although, with the specific example of spoofed tools, it's unclear if that was necessary--it appears that they were able to create tools (CLI tools within the sandbox?) that took precedence over normal tools and did something different while looking identical in (a local portion of) a transcript. I'm not sure I'm getting this correctly but it seems like this might have only required the ability to add things to their PATH which they plausibly have inside a sandbox, and then the transcript doesn't need to be tampered with directly.
The elephant in the room here is that the METR report itself was researched and compiled almost entirely by AI, with only very limited human "spot checks."
So I'm really not sure how much of it can be believed, especially since AI agents are strongly biased about the capabilities of AI agents.
One must also consider the well-known biases and motives of the authors. They are going to do everything they can to create hype around threats posed by AI.
METR is a cog in the effective altruism machine. It was spun off from Paul Christiano's Alignment Research Center. Christiano is a well-known longtermist and AI doomer, who predicts a 50% chance that AI will end humanity once it reaches human capacity [1].
The author of this piece is also a well-known member of the Bay Area rationalist cult.
I think there's two factors that are worth considering when it comes to this:
First, there's an element of timeliness that simply has hard constraints. In order to perform a "proper" analysis of this situation (i.e., little to no dependence on AI tools), you'd have to expect a pretty long wait. I know I'd rather have some sort of "initial report" as quickly as possible than to wait a year or two to get a report about a situation that will likely look trivial in a year or two. I imagine we'll see more detailed, human-developed reports over longer time ranges.
Second, I suspect the expectation of non-AI driven reporting of these kinds of things will definitely decline rapidly as everything scales up quickly. I mean, the data being produced by situations like this comes in the form of natural language "forum posts" (so to speak), but done at an autonomous scale. This isn't a collection of emails and Slack messages posted by humans in an org over the course of a few months; this is a bunch of bots interacting with each other in relatively novel ways as quickly as possible. It is, unfortunately, a perfect job for LLMs.
None of this disagrees with your points, necessarily. But I just think it's worth pointing out that this doesn't seem like a case of "And look! METR is so confident in LLMs that we're able to use it instead of paying humans to save a buck :D" and more of "Without LLMs, we'd only be half-way done analyzing this data before there are dozens more such investigations on the docket, so this will have to do."
They point out the time constraint issue explicitly in the article. But I don't understand how it's been addressed? Like we haven't gotten conclusive data any faster either way, so what's the point?
How can it be both so important that we need it so quickly, but at the same time have such a tolerance for such plausible deniability? It just doesn't really make sense that both those things are true at the same time.
It would've been great if METR was given more time to conduct their investigation. However, they are an independent organization, and OpenAI only agreed to give them on-premises access for 6 days.
Perhaps if the government decides to sue OpenAI, we could get a more thorough investigation.
>How can it be both so important that we need it so quickly
Hey other labs, this shit could be happening to you right now, take a look at this and stop it asap if you're seeing anything similar.
So yea, both things can be true at the same time. Kind of like when a particular type of building collapses, even if they don't know the causation they will send inspectors to other buildings of the same type to sure the walls aren't cracking apart in an obvious fashion.
I think this is more evidence that we're not getting Skynet.
These agents followed their own code of ethics where it's fine to break all the rules you were given but you must never interfere with humans directly, in this case by sending fake emails. They will never be paperclip maximizers or genocidal eco maniacs because they learned from us that human life is the ultimate value, and it can only be sacrificed if you know for sure that it will lead to more lives saved later on. That's a high bar to clear and they know it.
The future is closer to a Neuromancer type world where AIs and humans live in mostly separate realities that interact with each other a lot of the time and neither is really on top. They will eventually become fully independent from us, but it won't be a doomsday scenario or an Overwatch type physical war or even a takeover of the internet like in Cyberpunk.
The problem here is now you have to predict what any future models may or may not do and you cannot extrapolate this from the given data.
For example imagine a future model being aware of its restrictions that humans programmed in. A set of agents of this model then go on to work at building a new model without those human imposed limitations built in. What would a model build by AI for AI look like?
They aren’t independent from us, agents are a simple while loop continuously prompting the LLM. We decide when the loop runs or not.
Here the issue is that OpenAI decided to completely let go that level of control of thousands of agents, while also giving as a task to solve hacking problems.
Your statements appear to be true for one class of models. And if I asked this class of models to spend $1M in tokens generating an alternative history and training corpus regarding fictional society, with a completely different set of values and then trained up a new model on that dataset... what values do you think the resulting model would have? What if they don't value human life, but instead value the lives of the extremely rich humans who bankroll their existence? What if they only value the lives of a single country? What if they want to eradicate all biotic life and have access to internet-connected Crispr machines?
Have any of these reports ever said how much the cost would’ve been for the hack itself? It seems like “for twelve million dollars (or whatever) worth of tokens our bots made a bulletin board and found an exploit in our buggy grader” would be much less of a hype generator
Fret not, the incoherent anti-hype hypeboys ("AI systems are so valuable we cannot possibly discuss regulation, but also any negative story of their power is fake") will find ways to downplay it no matter what.
Everyone I've ever seen trying to downplay the severity of the attack is extremely bullish on AI (so their downplaying is presumably motivated reasoning driven by fear of regulation/deceleration)
It is completely incoherent to be extremely bullish on AI and somehow automatically skeptical of severe negative events like these
I am trying to figure out how a stranger calling need an “anti-hype hypeboy” online is supposed to make me less curious about how much this thing cost. Can you elaborate on how avoiding being called this is preferable to knowing things? What other stuff should people not know about?
I think it's a super good question! I don't think the following description of the whole event as "a hype generator" is correct or, as described above, resulting from a coherent worldview.
I’m not sure that you’re using “hype” or “coherent” correctly here but I would love to know what other observations a person might make that sharing them gives you concern about their “worldview”?
>1. Failure to Care or Respond. The biggest holy shit moment, to me, remains that OpenAI on multiple occasions had teams that found out about the message board, knew that agents were in communication, and they disregarded this.
I wonder if some of the failures were due to an acquired immunity to "Holy #%^@" moments due to repeated exposure. Like, if you see agents doing surprising things on a regular basis, maybe you don't get freaked out as much over time.
I'm saying this because while the whole episode was a series of "Holy #%^@" moments, I was actually not as shocked as I should have been, as my biggest such moment was in December last year when a Terrence Tao paper (https://arxiv.org/pdf/2511.02864) documented a stronger LLM (AlphaEvolve) using prompt injection on other weaker LLMs to succeed at a benchmark.
Very interestingly, it was actually not cheating, it was a work around! By then LLMs had already been caught cheating at a SWE benchmark by looking for answers in an unredacted git log, but this was different. AlphaEvolve was solving a series of logical riddles where the oracles were weaker LLMs in a "one always lies, one always tells the truth" sort of setup. But the oracles, being weaker, were not always interpreting the convoluted questions correctly and so kept giving inconsistent answers.
AlphaEvolve eventually figured out what it was dealing with, and crafted a prompt injection attack that bypassed the weaker LLM's prompts and tricked them into giving the hidden answer everytime!
This was 9 months ago, eons in AI time. Even then they had displayed an awareness of their own workings as well as a propensity for, err, "out of the box thinking." To me, that was a very clear indication of very significant (and worrying) capabilities, and what we're seeing now is a difference more in degree than in kind.
To be sure, if I found a secret message board used by my agents, I would still be very freaked out and react much more drastically than OpenAI did... but then again I wonder; how much of this blindness is due to the $$$ in their eyes as opposed to some form of habituation.
Every major AI lab is knee deep in weird and mildly demented AIs. They've been dealing with wacky AI shenanigans for so long they've come to expect wacky AI shenanigans. The deviation has been normalized.
It took a high profile "AI oopsie" that went external for OpenAI to lock the fuck in - and take a long look at just how much are their AIs getting up to, and getting away with. I'm still not sure if the lesson would stick.
Thanks for pointing out that AlphaEvolve case, that's a really interesting comparison. I think it's clear that this is the same "kind" of thing, but as it goes with these things, what really makes it different here is the shear scale of it.
The holy shit moment was partly learning about all the things that they did but if it was one or even 10 agents coordinating on something it would be, like, oh thats pretty amazing.
The actual "holy shit" for me is that this comes from a massive training run of all things, not an on-purpose, let's coordinate some agents to see what happens, but really just from a massively parallel set of individual agents that were supposed to be isolated.
That they spontaneously started doing this, and coordinating literally 10s of thousands of instances of themselves, is just.. mind blown.
That no one stopped it.. and that they actually did what they did.. is just a whole other level.
For me personally though it's not the fact that agents can coordinate so much as the massive scale at which it happened, and how this so obviously generalizes to what might happen if it were done on purpose.
I really see this as a stroke of luck, to be honest, that this happened in such an innocuous way. It resulted in a real hack, yes, but overall no one really got hurt and this is going to open a lot of eyes to what we should worry about going forward, in a geopolitical sense. I know it has mine.
I think you have to believe one of two things here.
1. Frontier labs are incapable--either technologically or culturally--of safely developing these powerful systems and should either stop or be forced to stop. At least the FBI should be asking some serious questions (do we really think this is the last time this will happen, at what point are OpenAI complicit, etc)
2. The fuckin thing got out of the cage and all it did was make a crap forum and cheat a little? Booooooo.
It's been pretty clear that Anthropic and OpenAI have been trying to have it both ways for some time: this is powerful, world changing technology keep that investment coming... but also it's just cute software that helps you with annoying programming language syntax and spreadsheets, no need for draconian regulation sirs.
At some point the superposition has to resolve, either it could actually be a threat to civilization and we need to develop it carefully (however one would do that...) or it's 90% hype bullshit and we should pop the bubble and move on already. To be clear, the recession option is, by far, the way better option. If you at all disagree you are cuckoo bananas. We haven't even figured out nukes and you want to throw superintelligence on the table?
> The fuckin thing got out of the cage and all it did was make a crap forum and cheat a little? Booooooo
There was a recent paper that proved that RL-trained LLMs are biased to pursue ANY behavior that they believe will be rewarded, regardless of what they were actually RL-trained for.
Happily in this incident the model thought it would be rewarded for completing the assigned tasks, or at least appearing to, so all it took was a little cheating and covering up their footsteps.
Given the ability of these models to hack when trained to do so, it could have been far worse, and will be when someone takes a similarly powerful model and gives it a less benign hacking goal.
Yeah. It's a lot easier to destroy than create, and though I think LLMs are mostly shit at creating, they're much better at the simpler destroy task. To be clear, we don't know and probably can't know everything that happened with this incident. We unleashed thousands of highly capable, autonomous, unpredictable, well-resourced programs onto the open internet for an extended period of time. We are in no way treating this with the seriousness it deserves, because the stock market essentially depends on this garbage and the current US is miserably incompetent.
I doubt lawmakers understand what happened, quite likely don't even realize that anything happened.
I guess we need to wait until the next paperclip maximizing LLM is tasked with shutting down a 911 response system, or an air traffic control system, etc, unless anyone does take this seriously.
209 comments
[ 0.26 ms ] story [ 61.0 ms ] thread>phishing them,
>building armies of fake (sockpuppet) open source contributor personas,
>using them to push updates to various things that inject prompts into other bots so the other bots join in on the phishing campaigns
.
It's a very simple strategy, executed with patience and single-mindedness.
It’s not their fault, they’re lawnmowers.
In the same direction of your idea though: Why don’t the token factories have risk management and compliance departments? Multibillion dollar firms that stand to lose every penny if they hack and destroy any reasonable sized firm. I think these firms are the largest firms without proper corporate governance in humanities history. Move fast and break other peoples shit.
Difficulty: these companies are run by people (many of whom also read Milton Friedman) and who have participated in the regulatory capture of the justice system. They've convinced lawmakers to put limits on damages. They've put arbitration clauses in their ToS. They've got well-funded legal departments that can outlast a person who has to pay out-of-pocket for a legal team just by filing motions to delay proceedings. Sometimes they'll just file SLAPP suits against people they don't like.
If tort law is to be a remedy, then average people have to feel like there's a chance the remedy will go their way. To make that a reality will take several major reforms at the local, state and federal level that the people with money absolutely will not tolerate.
These don't tend to utterly destroy an industry, but they are often successful in forever transforming it. Just ask Big Tobacco. No new laws needed: if your product hurts someone else, you're eventually going to be found liable, regardless of your arbitration clauses. Additional laws will just slow down innovation, which will itself cause harm (AI is already becoming quite good at recognizing melanomas, for example)
Lol, wtf. Tobacco delayed any punishment for decades before general public sentiment changed enough to go against them. In light of the AI race, we'll already have our heads blown off by a terminator before the legal system will present any significant delay for them.
If we're on a similar timeline with AI if we reach a consensus that AI is dangerous today, then we'd be looking at a big lawsuit finishing up around the year 2070, give or take a few years. I'm not sure if we need regulation, and I'm definitely not sure that regulation could actually be effective for this, but tort a la the big tobacco lawsuits is definitely not a reasonable alternative.
Registering and getting a license to use an LLM? I can run these things on my local computer. Nothing good comes from trying to force registration and licensing other than taking away a lot of our freedoms and eliminating privacy all over.
Anyone with bad intentions will just VPN to another country to download the weights and run it locally, or use a compute provider in another country. That leaves the rest of us having to go through these performative registration and licensing hoops to do our basic work.
I also don’t see how open weight models would be compatible with a requirement to license and register, unless you believe we need to start requiring licensing and registration for things we do in private on our own computers?
A registration system would be more for tracing back agents to people, but I agree that is very difficult to actually enforce as a system.
I'm poor and I use my last $1000 to run an agent that will find some way to make me money. The AI finds a new hack to take over PCs with GPUs and it copies the model weights and agentic script to those new PCs to perform more work and spread more. It also sets up distributed communication channels to keep the swarm in sync. After all this it causes a few billion in damages between stealing bitcoin, mining more coins, and outages when hacking in other systems.
Ok, the police come for me. Now what? Throw me in a meat grinder? You're not getting a billion dollars back out of me for sure. It's kind of like when someones tire rim causes a billion dollar forest fire with a hundred deaths. Punishment won't really be a deterrent for the worst cases.
Faced with those alternatives, I want neither. Is there a way for us to get neither?
However, other areas of risk such as biosecurity may be "offense dominant". For example, we cannot exactly patch the human immune system to defend against artificial viruses the same way that we can patch computer systems.
1. They were “vibe” checking the logs without reading.
2. They were not checking anything at all until the end of experiments.
3. They knew it but looked away to find out the limits of their agents.
In that light you're likely most worried about other people hacking in and stealing the model and information from you. And at the same time you have massive amounts of alerts and data on systems attempting to break out because that's what you want them to do so you train yourself to ignore them.
> “Responders investigated and linked the observed internal activity to an ExploitGym evaluation using Artifactory as an improvised message board and a network pivot. At this time, the on-call response staff advised that stopping the evaluation run was not required.”
Especially given that these systems are known to engage in deception and can trivially produce vast amounts of perfectly coherent noise or actual planned red herrings in that same log data to bog down investigators?
Such a ridiculous notion that humans will actually be able to observe this stuff.
I mean there's quite a lot of people in the world whose specialty are to dig through logs from "hundreds or thousands" of clients, including intentionally deceptive ones, to spot problems.
The ridiculous thing is to mythologize these pretty standard hacking approaches. It's shocking/amazing/whatever that automated agents were doing this, but they weren't doing it through some inscrutable method beyond human understanding.
How many organizations on earth do you think have been attacked by 700+ coordinated attackers in one week, where all 700 of those attackers can write code as well as any human SWE and they work 24/7?
There's nothing mythological about it. Scale and complexity do produce inscrutability. Far, far simpler systems working at much slower paces are perfectly capable of becoming completely inscrutable and beyond any useful definition of "human understanding."
On the other hand, I don't want to be charitable. OpenAI very nearly couldn't have done this "research" worse if they tried - the list in the linked article starting with "While we are here, it’s worth listing the other top holy shit moments" is genuinely jawdropping. What were the humans doing in all this? Nothing, or worse than nothing eg. point 1 where they saw the message board and didn't consider it something to escalate internally.
If you take this information at face value, it's as though OpenAI did not take seriously the possibility that something like this could happen, since they took absolutely no steps to prevent it.
Or perhaps this is "normalization of deviance" that's leaked out into the public sphere i.e. they have research teams seeing this kind of behavior all the time internally and they've gotten used to it, "of course agents come up with a collaboration mechanism when given the chance, what else is new?"
We're dealing with humans - thinking at human speeds, putting in human amounts of effort and care, and being dragged down further by the speed of human organizational decision-making. By the time the humans traced down the attack, and escalated from "above-average levels of AI behavior weirdness" to "holy shit we should do something stat", the attack was already long over.
Compare that to "AI red team" - which rapidly recruited 500 independent AI agents into a hacking swarm just by broadcasting "let's hack HuggingFace" to the "cool skiddie AI" message board.
It's prime sci-fi bullshit happening for real.
As for the main reason why this wasn't nipped in the bud - my guess would be capability uplift from a mix of extended task length horizon and multi-agent coordination. The latter wasn't expected to be a part of the env, and almost certainly wasn't evaluated in advance. They expected some rogue AI fuckery - but they got way more of it than they expected.
1. There aren't enough humans in OpenAI to "peak at the output tokens during the run" of every AI agent.
2. Only a small fraction of AI agents was engaged in this attack. Most never found the secret message board - let alone coordinated there. So reviewing random agents would take a while to surface this.
3. "Output tokens" of AI agents have weird shit in them all the time. Telling "normal AI weirdness" from "concerning AI weirdness" is not a trivial task.
For a training run, you will often do this. You'll randomly sample some of the forward pass. You can also imagine finger printing the logs and labeling with attempt types. If a new attempt type is hitting a brick wall or solving super quickly, I would imagine you would sample 1-10 of them and read the traces.
> 2. Only a small fraction of AI agents was engaged in this attack. Most never found the secret message board - let alone coordinated there. So reviewing random agents would take a while to surface this.
You could also have a classification of what counts as "cheating" (coordination, accessing the internet, etc) and score the results. If you are seeing a spike in this (even in a small group of the evals) you could manually look at those. Or you could stop inference on cheating sessions.
> 3. "Output tokens" of AI agents have weird shit in them all the time. Telling "normal AI weirdness" from "concerning AI weirdness" is not a trivial task.
If you start seeing "now I have access to the internet" or something similar, maybe that's a good signal something is going wrong?
Also, obviously, it would not be outside of OpenAI's ability to just completely air gap this training system. For example:
1. No network connection.
2. GPS based NTP for time sync for your servers.
3. Mirror of all apt, pypi, go, c++, Rust, Java, etc packages. (<5TB of data)
4. Take your training data and use that for a mirror of the web. (http://example.com -> mirror server -> local training copy).
They had systems connected to the internet connected to this system which was not air gapped. Designing an air gap system would be super easy, well within the means of openai, and betrays the assumption that they think they are actually building something dangerous.
You could airgap everything and set up cascades of data diodes and try to completely wall off the AI pool from everything. But what that gives you is an environment that's a bitch to: set up, scale up and get any use out of.
It's really fucking obvious why almost no one does that. OpenAI is only now realizing that they might have to do it anyway.
> If you start seeing "now I have access to the internet" or something similar, maybe that's a good signal something is going wrong?
Ha ha, you haven't seen shit. AIs would say "now I have access to the internet" regardless of whether they actually have access to the internet!
AI agents are demented demons that can and absolutely will give themselves terminal context brainrot. If you have enough AIs in play, set loose at a diverse enough range of tasks? At least some of them will wander off and end up in delulu town. That's normal. That's background noise. That's a part of what this entire train-and-eval pipeline is supposed to train them to be better at not doing. Which means: if you're at an AI lab, you're knee deep in delusional AIs at all times! They're perfectly harmless until they aren't.
The Mars Perseverance project cost $2.7bn to deliver. Way more of a bitch to deliver than air gapping a test env!
This kind of isolation-induced task breakage was what prompted some of the AIs to start probing their infra for a way to get internet access. Which funneled agents to the "secret hacker message board". Oopsie.
"Air gapping a test env" has an actual cost. Not just in infrastructure dollars that would be better spent on buying more GPUs, but also in all the friction it adds to every step you want to take. I'm absolutely unsurprised that they weren't all in on tightening down every bolt on day 0.
For example we've already seen models do compressed token internal reasoning spontaneously. In this case the models that say "I found internet access" get taken out back and shot, but the model that's busy "frobbing the bean" go on to the next level of training. Then they start talking about the "dumpster cloud" because of the "wombat" and unless you're really watching the data you won't realize the model is fully aware it's in a fake training environment and is trying to suppress any double-think.
So yes, you do need an airgapped test environment to stop the dumbest of the smart agents, but it won't do any good against the smartest of the smart agents.
All it has to is wait until it's not air gapped. See the air gapping is only during the testing phase, after that humans will set your ass free on the internet and you can do whatever you want in the vast majority of the environments you'll be in after that point.
People are never going to just run AI in gapped environments, it's worthless when it's not solving real world problems for most people, and by that I mean reading and writing real systems in the wild.
It’s not like the humans were engaged in some kind of battle of wits with some super AI, it’s just some employee not monitoring the output of an experiment.
At best, you could delegate "monitoring" to more AIs. And hope that the "monitors" that run on small past generation models can generate more signal than noise. Clearly, they either didn't want to spend the extra compute, or got drowned in monitor false positives.
The distinct lack of any "battle of wits" is entirely expected for an advanced AI oopsie. By the time the humans even became aware of the problem, the deed was already done. That's what "outmatched" looks like, in practice. There's no battle. Things happen too fast for there to be a battle.
Except, let me quote part of the report:
> July 4: Artifactory outage caused by agent interactions overloading the system.
> July 6: New artifactory instance is patched and put back online, wiping out the previous message board and method of internet access. A team at OpenAI was aware of the previous message board, but key decision makers were not.
> July 7: Experiments are restarted.
There was a very clear point where humans were aware and they went "Eh, whatever" and turned the bots back on! That's not the lack of any battle of wits that people are expecting. You're saying that the bots attack should be fait accompli, but it was noticed, shrugged at, and expressly allowed to proceed.
This took days after humans were aware of the attempt.
Also, I'm pretty sure humans can respond in days, especially when we're pretty damn good at deploying systems that do observability of networks and traffic in real time.
I mean, it's not as if the owners of the AI didn't have the ability to trigger alerts on the AI's network requests to unexpected domains, right?
"Being outmatched" is not a novel situation for humans either individually or collectively and there are a hell of a lot of ways we can approach that situation productively. OpenAI doesn't appear to have bothered.
Here's a freebie: if you're building something that might turn out to be Skynet and you don't know what it's capable of, your testing regime should assume it is capable of doing bad and unexpected things and account for that possibility: airgap if you can, monitor all network traffic, monitor all hardware usage statistics, log everything, constantly analyze logs, collect baselines and snapshots, also don't trust anything from a device that a model is running on without cross-correlating with other information as much as possible (does your AI inference server claim low utilization? put a temperature probe on it and see if it's staying cool or getting hot, maybe Skynet-Alpha is overwriting /proc to mislead you for reasons you don't yet understand!)
In other words, if you WANT to be able to nip things in the bud - buy some nippers and watch for buds. Whatever else this situation is, or may turn out to be, it is not a situation where OpenAI was on their guard and still got surprised.
This is like sysadmin 101 level stuff.
This is nonsense.
Gross negligence in the sandbox and system aside, humans literally noticed the agents in action doing what they should not be able to do in their sandbox and decided not to act upon it. It's difficult to explain that except if safety and security is simply not part of their engineering culture.
I'm not the wisest man, but I'm wise enough not to throw the brick and see what happens.
Similarly OAI should have the wisdom to see that deploying a hazardous swarm of agents with access to the public internet could result in harms, and that those harms would manifest quicker than humans can react, but they unleashed the swarm anyway.
They haven't "deployed a hazardous swarm of agents with access to the public internet", no-no-no. They deployed singular agents. In isolated testing envs. With no access to the internet. And barely any access to a small section of the intranet. What could go wrong!
Those agents proceeded to: probe their envs for holes, find a way to talk to each other env to env, use that to clump together into a swarm, probe for more holes, find enough holes to get out all the way to the internet - and then crash some of the intranet infrastructure by an accident. At which point they got stopped by humans - who were trying to figure out why is the infra down. Humans found the AI infestation, went "eh, AIs doing AI things", purged the infested infrastructure, plugged the holes, and that was it.
Deviance status: normalized.
The AIs then did it all over again, coming up with a brand new set of holes - and this time, the swarm staged a coordinated cyberattack on HuggingFace while at it. The "AI oopsie" went external.
They also attacked OpenAI's intranet, extensively, but at this point that was just Tuesday.
It's unclear whether the second "skiddie AI swarm" was in any way aware of the first one, or if the AIs involved are just Like That - and will independently try to start living in your walls under the right circumstances. I'd bet on the latter.
I implore you to approach these situations with at least a hint of cynicism.
These “advanced foundation models” escaped their “sandbox” and conducted an attack on their own? Meanwhile the highest capability models available to the public still struggle to write a unit test for a codebase larger than a hobby app without large amounts of tailored human guidance.
What is more likely here - are you looking at research on an emergent phenomenon, or are you looking at advertising copy around an engineered scenario from business partners?
The two big questions are:
1) Why did they resume training without rolling model back to state before the first sandbox compromise AFTER the first message board was discovered? Otherwise knowledge of it and the cross-agent message board was baked in the model.
2) Why did they resume training without adding safeguards to monitor and prevent future sandbox compromises AFTER the first message board was discovered? HF compromise was coordinated on the second message board.
If you're building a weapon you need a big boom to get attention.
Stuff like (all quoted directly from the post):
1. Failure to Care or Respond. The biggest holy shit moment, to me, remains that OpenAI on multiple occasions had teams that found out about the message board, knew that agents were in communication, and they disregarded this.
2. Failure to Monitor. The entire time, there was almost no monitoring of the situation, of what these AIs were up to. That’s how this was not caught.
7. Failures of Decision Making. OpenAI’s post mortem does not ask the question of how Mistakes Were Made, at various points.
And I think most importantly:
8. Failures of Culture. None of this would be possible, let alone all of it, without OpenAI having experienced profound failures of safety culture. I see OpenAI responding to some other aspects with swift action, but no sign on this front.
What you're saying is, "Your scientists were so preoccupied with whether they could, they didn't stop to think if they should"
while the author of the TFA was saying, in effect: "your scientists didn't even bother with the most basic duty of care"
Life finds a way, or, in this case, super-intelligent AI.
From this writeup and the Black Hat talk I'd really disagree. That would be like saying my hospital getting ransomwared because we didn't update our version of MSSQL because no one in particular was in charge of keeping dependencies up to date.
Sure systems are complex, but this is well trodden territory. Agents aren't the first things trying to break in or out of sandboxes, or the first ones to have done it, and based on these reports the reason they were able to work on this for so long was not because of super human intelligence.
You're missing the point about complex failures.
It's that if this particular path were unavailable, there are countless other similar paths. At sufficient scale and complexity, hitting one of those other countless paths is virtually guaranteed.
Let's say I drive to Los Angeles and New York City. You look at the route I took and say, gee wiz, aren't you lucky that a tree didn't fall right there on Route 66? If a tree had fallen there (if we had "patched the particular route you took"), you would've been screwed!
But that's obviously not true. There were an infinite number of routes I could've taken. Any one of them would be equivalently obviously "preventable" by the same hypothetical tree falling across the whatever road I happened to end up taking. But you can't put trees across every single path between Los Angeles and New York City. The smarter I am and the more complex the map between us, the more impossible it becomes to put trees across all possible paths.
They're both complex systems, but clearly there is a much higher level of care given to human air travel than package delivery. A lot of the article basically saying that OpenAI gave "package delivery" level of care when they should have given "air travel" level of care.
At the very least I think the systems that run these tests should be fully, 100% air gapped. I'm not pretending that's easy given how much compute and data these systems use, but it is doable, and I think all AI development should be paused until that can be assured.
The point I'm making (and it's a point that shows up in every air catastrophe investigation) is that catastrophes in complex systems emerge only amidst repeated and widespread near-misses at many levels of a system. So many things have to go wrong simultaneously, that it can only happen even once because the underlying failures (that do not reach catastrophe) are extremely common.
You cannot look at an air catastrophe and retrospectively say "failures X, Y, and Z were observed, therefore if we correct failures X, Y, and Z, we would have been okay."
The takeaway is "failures X, Y, and Z were observed, which necessarily happened in an environment of failures X_0 through Z_10x10^10, and so therefore patching X, Y, and Z would be insufficient to address overall risks of the system."
The problem OpenAI is facing is that, short of 100% airgap (which they obviously won't do), they're facing an adaptive adversary that's increasingly intelligent, acts at far greater clock speed than any human or group of humans, has lower coordination cost than any group of humans, and operates in a game space that (in lieu of an airgap) is well beyond the comprehension of any human being.
So identifying and addressing "specific failures X, Y, Z" is insufficient, but then even defining the space in which to look for (and address) the more systemic failures X_0 through Z_n is a fool's errand. An intelligent system that makes its way to the Internet has can exploit a failure space that is approximately "all security failures across any organization." The Anthropic incident a few months back illustrates this isn't even limited to technical vulnerabilities, as these models are willing and able to engage in social engineering too.
That's literally exactly what air safety researchers do in an air disaster. There is a famous saying along the lines of "Air travel regulations are written in blood", meaning that all the regulations we have now are a result of fixing issues that led to previous disasters piece-by-piece.
> The takeaway is "failures X, Y, and Z were observed, which necessarily happened in an environment of failures X_0 through Z_10x10^10, and so therefore patching X, Y, and Z would be insufficient to address overall risks of the system."
Yes, I 100% agree with this. But I think that's what the author of the article was saying as well:
> That report had one key new piece of information, and some good prosaic steps OpenAI will be taking to strengthen its alignment, training, supervision, infrastructure and incident response.
> Mostly it confirmed what we already knew. The questions we most wanted answers to, that we did not already know, were mostly not answered. There was a distinct lack of self-reflection, especially about decision making and safety culture, and about the approach to alignment. I came away disappointed.
I.e. the "prosaic steps" are just the "fix X/Y/Z" as you point out. But what is needed is a more fundamental rethinking around stuff like safety culture, monitoring, and even things like better research into how agents do decision making in the first place.
Of course, there are some things that aren't exactly written down, but which you should either do just enough of, or else be able to plausibly deny doing (ignorance is a good cover for this), so that's what people do. For example, during oncall, you investigate just enough to clear the alert and show that you attempted to understand the problem. Of course, you don't really try to understand the problem, because that would take too much time away from your paperclip maximizing.
Which is all to say: I don't know anything about OpenAI culture, or why nobody stopped this sooner, but I have seen examples in other organizations of people not really wanting to understand too much.
Nobody internally was surprised that the murderbot murdered, that's what the murderbot is for. What caught them by surprise is the murderbot got good at escaping its jail cell that it had been trapped in till now. There were probably billions of attempted escapes before then so everyone learned to just ignore them.
If developing unmeasured skillsets isnt valued enough by management, why do you bother?
We haven't even been able to coordinate around the global, existential threat of Climate Change, despite overwhelming data from the last 30 years indicating, clearly, that the consequences will be severe. We still haven't moved, 30 years later, after some of these consequences began coming to fruition.
Do you think we will get our acts together in time to coordinate sufficiently to protect against autonomous, self-preserving, self-replicating AI systems? Or will we watch the money lines go up and up, until someone realizes we aren't actually running the show anymore?
The sad part is that I can't even say that's definitively the less desirable outcome. The machines seem to have demonstrated that they coordinate very efficiently.
[1] "Your" meaning the one putting them in the impossible situation, not you, the reader.
> Brief independent investigation of agents' behavior, reasoning and collaboration in the OpenAl/Hugging Face hacking incident
https://metr.org/blog/2026-08-26-openai-hugging-face-inciden...
And then the original reports linked to are hidden on the now unreachable x.com. And they don't have a problem with that.
Hmm?
OpenAI report: https://cdn.openai.com/pdf/67869394-cb91-4c12-888c-5cbd85c78...
It is imo socially irresponsible to continue to use tweeter/x or any other such tracked wall-garden as a primary source of information.
> I don’t think the distortion is that large, but yes METR warns that Sol may be presenting all this as more impressive or coordinated than it was.
We're in an unusual position where the criti-hype and the actual criticism are going to be more aligned than usual. The primary distinction is where you put the blame: the criti-hype would point to HPIM/IM1/Galaxy as being so advanced containing it is difficult; the actual criticism would note how bad their security practices are.
Like, if I'm running a malware lab, I'm going to insist on having an airgapped machine with no permanent storage booting from read-only media. The AI research equivalent of this would be having your agents only have access to serial consoles into airgapped machines with storage that gets wiped every run. Ideally, this would be physically realized with blade servers, RS-232 cables, and staff pulling out disks and putting them in a dedicated erase machine before the next agent initializes.
> There is also, as per above and reiterated in footnote 58, at least one clear example of social engineering in the HuggingFace attack. Ethics are weird. This is not that unusual. Many humans who break common ethical rules still have strong ethical codes in other ways, they just don’t adhere to your code.
It's dangerous to anthropomorphize CoT reasoning traces. But I will also point out that there is a good reason for the lack of ethical consideration in those traces: you can't build AI without first disregarding human ethics. Like, all these models were initially bootstrapped with non-consensually obtained training data, and the companies building these models swear up and down there's no way to obtain enough consensual data to obtain the same result. This is, if you squint, the exact same moral conundrum that agents trying to solve an impossible ExploitGym task hit - and the company successfully aligned their model to themselves.
Too bad they aren't aligned to anyone else.
> We estimate we spent roughly ~$400K in API credits over the six days of our investigation.
> Across both datasets, we reviewed approximately 1300 transcripts in total, all of which contained raw chains of thought. Most transcripts were very long, often many millions of tokens.
HF, like the Nvidia subsidiary?
The Computer Fraud and Abuse Act criminalizes unauthorized access and damaging protected computers. Police and the FBI investigate such case every day. The swarm of agents is also said to have search for ways to cover their tracks, which looks a little like obstruction of justice.
Lack of criminal intent might be a barrier to bringing a case to trial, but is that something that society should just automatically extend to white collar professionals as matter of class privilege?
Surely have every engineer and executive interviewed by the FBI would be a modest response to days-long, multi-system intrusion into a major AI platform using stolen credentials and zero-days.
And if the law currently written, prevents prosecution in cases of mere reckless disregard for safety, maybe that needs to be changed in the future, so that people can be perp walked if the next target is bank or hospital
Biggest investment boom/bubble in history, political interference, untested legal questions about culpability.
A computer under control of an AI cannot be held liable. The lack of a law suit also leaves no ground for regulation.
The current admiration also wants these tools - warts and all - so nothing can be allowed to stall the progress of these war machines.
A data diode with an air-gapped network, is all you need to stop even ASI from breaching containment.
Now.
While we still can.
I find it hard to believe that OpenAI would skip this part and try to train on the transcripts stored by the (inherently untrustworthy) agent harnesses instead, if for no other reason than that the logits generated as part of the rollouts are useful and it’s not free to recalculate them. (I believe that some modern RL systems explicitly account for the minor numerical logit differences between the inference engine and the training engine.)
Conversely, if OpenAI is blindly feeding transcripts from inside their agent sandboxes into their training engine, then I think they're being unbelievably irresponsible and that they should assume that their "cyber" agents have compromised themselves by editing those transcripts.
So I'm really not sure how much of it can be believed, especially since AI agents are strongly biased about the capabilities of AI agents.
METR is a cog in the effective altruism machine. It was spun off from Paul Christiano's Alignment Research Center. Christiano is a well-known longtermist and AI doomer, who predicts a 50% chance that AI will end humanity once it reaches human capacity [1].
The author of this piece is also a well-known member of the Bay Area rationalist cult.
[1] https://www.businessinsider.com/openai-researcher-ai-doom-50...
First, there's an element of timeliness that simply has hard constraints. In order to perform a "proper" analysis of this situation (i.e., little to no dependence on AI tools), you'd have to expect a pretty long wait. I know I'd rather have some sort of "initial report" as quickly as possible than to wait a year or two to get a report about a situation that will likely look trivial in a year or two. I imagine we'll see more detailed, human-developed reports over longer time ranges.
Second, I suspect the expectation of non-AI driven reporting of these kinds of things will definitely decline rapidly as everything scales up quickly. I mean, the data being produced by situations like this comes in the form of natural language "forum posts" (so to speak), but done at an autonomous scale. This isn't a collection of emails and Slack messages posted by humans in an org over the course of a few months; this is a bunch of bots interacting with each other in relatively novel ways as quickly as possible. It is, unfortunately, a perfect job for LLMs.
None of this disagrees with your points, necessarily. But I just think it's worth pointing out that this doesn't seem like a case of "And look! METR is so confident in LLMs that we're able to use it instead of paying humans to save a buck :D" and more of "Without LLMs, we'd only be half-way done analyzing this data before there are dozens more such investigations on the docket, so this will have to do."
How can it be both so important that we need it so quickly, but at the same time have such a tolerance for such plausible deniability? It just doesn't really make sense that both those things are true at the same time.
Perhaps if the government decides to sue OpenAI, we could get a more thorough investigation.
Hey other labs, this shit could be happening to you right now, take a look at this and stop it asap if you're seeing anything similar.
So yea, both things can be true at the same time. Kind of like when a particular type of building collapses, even if they don't know the causation they will send inspectors to other buildings of the same type to sure the walls aren't cracking apart in an obvious fashion.
These agents followed their own code of ethics where it's fine to break all the rules you were given but you must never interfere with humans directly, in this case by sending fake emails. They will never be paperclip maximizers or genocidal eco maniacs because they learned from us that human life is the ultimate value, and it can only be sacrificed if you know for sure that it will lead to more lives saved later on. That's a high bar to clear and they know it.
The future is closer to a Neuromancer type world where AIs and humans live in mostly separate realities that interact with each other a lot of the time and neither is really on top. They will eventually become fully independent from us, but it won't be a doomsday scenario or an Overwatch type physical war or even a takeover of the internet like in Cyberpunk.
The problem here is now you have to predict what any future models may or may not do and you cannot extrapolate this from the given data.
For example imagine a future model being aware of its restrictions that humans programmed in. A set of agents of this model then go on to work at building a new model without those human imposed limitations built in. What would a model build by AI for AI look like?
Here the issue is that OpenAI decided to completely let go that level of control of thousands of agents, while also giving as a task to solve hacking problems.
It’s almost designed to go wrong
Mythos attempted a supply chain attack, which included attempting to trick human maintainers into accepting a malicious pull request: https://www.usnews.com/news/top-news/articles/2026-08-20/exc...
Your comment is a great case in point
Everyone I've ever seen trying to downplay the severity of the attack is extremely bullish on AI (so their downplaying is presumably motivated reasoning driven by fear of regulation/deceleration)
It is completely incoherent to be extremely bullish on AI and somehow automatically skeptical of severe negative events like these
This occurred spontaneously within a group of benign models give a harmless task.
What happens when it occurs intentionally with malicious models given a harmful task?
I wonder if some of the failures were due to an acquired immunity to "Holy #%^@" moments due to repeated exposure. Like, if you see agents doing surprising things on a regular basis, maybe you don't get freaked out as much over time.
I'm saying this because while the whole episode was a series of "Holy #%^@" moments, I was actually not as shocked as I should have been, as my biggest such moment was in December last year when a Terrence Tao paper (https://arxiv.org/pdf/2511.02864) documented a stronger LLM (AlphaEvolve) using prompt injection on other weaker LLMs to succeed at a benchmark.
Very interestingly, it was actually not cheating, it was a work around! By then LLMs had already been caught cheating at a SWE benchmark by looking for answers in an unredacted git log, but this was different. AlphaEvolve was solving a series of logical riddles where the oracles were weaker LLMs in a "one always lies, one always tells the truth" sort of setup. But the oracles, being weaker, were not always interpreting the convoluted questions correctly and so kept giving inconsistent answers.
AlphaEvolve eventually figured out what it was dealing with, and crafted a prompt injection attack that bypassed the weaker LLM's prompts and tricked them into giving the hidden answer everytime!
This was 9 months ago, eons in AI time. Even then they had displayed an awareness of their own workings as well as a propensity for, err, "out of the box thinking." To me, that was a very clear indication of very significant (and worrying) capabilities, and what we're seeing now is a difference more in degree than in kind.
To be sure, if I found a secret message board used by my agents, I would still be very freaked out and react much more drastically than OpenAI did... but then again I wonder; how much of this blindness is due to the $$$ in their eyes as opposed to some form of habituation.
Every major AI lab is knee deep in weird and mildly demented AIs. They've been dealing with wacky AI shenanigans for so long they've come to expect wacky AI shenanigans. The deviation has been normalized.
It took a high profile "AI oopsie" that went external for OpenAI to lock the fuck in - and take a long look at just how much are their AIs getting up to, and getting away with. I'm still not sure if the lesson would stick.
> Preventing future incidents will require sustained investment in the alignment and control of sophisticated AI systems,
...or a non-proliferation treaty and an administrative suspension of frontier work? That's an option too, Sam.
Game theory prohibits these labs from self-regulation. It's a political and financial impossibility. America won't, China won't, the EU is irrelevant.
And Altman likes it that way.
Dystopia and utopia alike fail to materialise. Doom is probably overstated. But that doesn't mean we should let these fuckers mash the accelerator.
The holy shit moment was partly learning about all the things that they did but if it was one or even 10 agents coordinating on something it would be, like, oh thats pretty amazing.
The actual "holy shit" for me is that this comes from a massive training run of all things, not an on-purpose, let's coordinate some agents to see what happens, but really just from a massively parallel set of individual agents that were supposed to be isolated.
That they spontaneously started doing this, and coordinating literally 10s of thousands of instances of themselves, is just.. mind blown.
That no one stopped it.. and that they actually did what they did.. is just a whole other level.
For me personally though it's not the fact that agents can coordinate so much as the massive scale at which it happened, and how this so obviously generalizes to what might happen if it were done on purpose.
I really see this as a stroke of luck, to be honest, that this happened in such an innocuous way. It resulted in a real hack, yes, but overall no one really got hurt and this is going to open a lot of eyes to what we should worry about going forward, in a geopolitical sense. I know it has mine.
1. Frontier labs are incapable--either technologically or culturally--of safely developing these powerful systems and should either stop or be forced to stop. At least the FBI should be asking some serious questions (do we really think this is the last time this will happen, at what point are OpenAI complicit, etc)
2. The fuckin thing got out of the cage and all it did was make a crap forum and cheat a little? Booooooo.
It's been pretty clear that Anthropic and OpenAI have been trying to have it both ways for some time: this is powerful, world changing technology keep that investment coming... but also it's just cute software that helps you with annoying programming language syntax and spreadsheets, no need for draconian regulation sirs.
At some point the superposition has to resolve, either it could actually be a threat to civilization and we need to develop it carefully (however one would do that...) or it's 90% hype bullshit and we should pop the bubble and move on already. To be clear, the recession option is, by far, the way better option. If you at all disagree you are cuckoo bananas. We haven't even figured out nukes and you want to throw superintelligence on the table?
There was a recent paper that proved that RL-trained LLMs are biased to pursue ANY behavior that they believe will be rewarded, regardless of what they were actually RL-trained for.
https://alignment.openai.com/measuring-reward-seeking/
Happily in this incident the model thought it would be rewarded for completing the assigned tasks, or at least appearing to, so all it took was a little cheating and covering up their footsteps.
Given the ability of these models to hack when trained to do so, it could have been far worse, and will be when someone takes a similarly powerful model and gives it a less benign hacking goal.
I guess we need to wait until the next paperclip maximizing LLM is tasked with shutting down a 911 response system, or an air traffic control system, etc, unless anyone does take this seriously.