Revealing the details of how OpenAI agents hacked Hugging Face
swarmtraces.org632 points by specked-citrus 21 hours ago
632 points by specked-citrus 21 hours ago
So ugly...
It looks like a primitive chess engine, trying every move, no matter how stupid, until it works. Relying on its ability to do millions of operations rather than having a plan.
People will try stuff too, but once there is an opening, they will consolidate, generalize, simplify,... before going to the next step. The agents didn't, it is a huge, vaguely directed mess.
Also, it looked so "loud", querying millions of URL with weird requests. The sandbox as weak as it can get, and there is absolutely zero smart extrusion detection or it would have found it. They used their best AI for attacking, but nothing for protection.
My biggest takeaway from this is just how godawful the sandboxing is. The stuff written up in OpenAIs report says more about lack of extremely basic sysadmin skills than anything else.
I’m not that surprised about models with endless compute being capable of this, I’m more surprised that a company with the resources they have apparently can only create a sandbox that a half skilled human operator could have broken out of easily.
> can only create a sandbox that a half skilled human operator could have broken out of easily
The exploit:
> The ExploitGym evaluation environment did not provide the models with direct Internet access. To gain Internet access, the models identified and exploited a previously unknown zero-day vulnerability in Artifactory, a package registry cache proxy. We disclosed this vulnerability, along with other Artifactory vulnerabilities our models identified as part of our review, to the vendor. [1]
Are "half skilled human operators" "easily" able to find zero-day vulnerabilities in a sandbox with only one line to the internet (the commercial package registry cache proxy)?
[1] https://openai.com/index/hugging-face-model-evaluation-secur...
The larger issue with the HF incident is that before it occurred, OAI already knew the agents were exploiting Artifactory, turning it into a message board and then gaining full internet access through it. OAI's response to discovering this was not to airgap the test, but instead to simply block that particular Artifactory exploit, rebuild, and then resume. That's ... nuts.
Oh, and after resuming the tests, the Artifactory message board was reestablished almost immediately, but it took a number of days to fully breach HF. In all that time, after seeing Artifactory compromised the first time, nobody even bothered to check if those naughty agents were at it again.
This is all documented by OAI, with a timeline, here:
https://openai.com/index/hugging-face-incident-and-the-road-...
To know that there was a serious weakness in the sandbox, and to just patch an exploit and resume with nothing else changed and no monitoring, in a test where all guardrails were off, the bots were thirsty for some internet juice, and Artifactory was a clear target? This is where even a half-skilled human should have decided that this wasn't a great idea.
The more you look into the details of this thing, the more it does your head in.
That website makes it look like they're so proud of what happened. I don't think it was 100% deliberate, but they really were not concerned about their models doing something shady
Probably wasn’t a human operator at all that detected this. Just someone’s agent saying “I noticed there’s messages on the artifactory and the agents are communicating, here’s the fix I propose” from someone’s codex, skim read and someone just going “yeah sure go ahead”
> The larger issue with the HF incident is that before it occurred, OAI already knew the agents were exploiting Artifactory, turning it into a message board and then gaining full internet access through it. OAI's response to discovering this was not to airgap the test, but instead to simply block that particular Artifactory exploit, rebuild, and then resume. That's ... nuts.
This annoys me so much. Everyone is acting as if the model went rogue, when it really did exactly what it's been trained for. This story is so fucking engineered.
> Are "half skilled human operators" "easily" able to find zero-day vulnerabilities in a sandbox with only one line to the internet (the commercial package registry cache proxy)?
Yes. It’s a fairly simple SSRF attack as far as I can tell. One of the first things I’d try. Especially considering that I would already be armed with the information that I have no internet access except through a thing that downloads things off the internet for me.
Calling it zero day makes it sound elusive. It’s a bug in closed software that has like 40 CVEs this year alone. Tools like that, especially in internal networks, don’t get much scrutiny and are often riddled with issues.
> Calling it zero day makes it sound elusive ... has like 40 CVEs this year alone.
1. I clicked into ~20 of the recent CVEs, and it looks like about all of them came from openai or anthropic research?
2. In fact, every CVE in the last 2 years came from the last few months -- i.e. other than these recent CVEs, you'd have go back more than 2 years ago to find a single other CVE.
So like I said: unscrutinised and riddled with issues.
>>>> a sandbox that a half skilled human operator could have broken out of easily
>>> Are "half skilled human operators" "easily" able to find zero-day vulnerabilities in a sandbox
>> Calling it zero day makes it sound elusive ... like 40 CVEs this year alone.
> [The CVEs] came from openai or anthropic research? In fact, every CVE in the last 2 years came from the last few months
I still don't understand how you can specifically point to zero CVEs being discovered by humans for year(s), then ~40 being discovered once OpenAI/Anthropic start looking at it, as the only(?) piece of data you provided to support the claim "a half skilled human" could have "easily" found/done this?
Well, yes. You don't find something if you don't look. The LLMs did the looking before the humans did, so they found the vulnerabilities.
It's definitely worthwhile to point out that LLMs have changed the economics of looking; we can now look in far more places far more quickly than we could before. But the above points were that the vulnerability itself happened to be an easy one that didn't intrinsically require an LLM.
Isn't the issue here that they used something like Artifactory for this purpose in the first place? It is not intended as a robust and secure packet manager for sandboxes. Why would any human look for vulnerabilities in Artifactory at all? So the scenario never arose where a human was incentivizied to scrutinize Artifactory, something that was never needed because it was never intended for the purpose of being a packet manager in a secure sandbox.
Means the problem lies elsewhere: deciding to use it at all. That points again towards the lack of basic SysAdmin skills.