Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Discovery of a new OpenAI agent message board

collusion.wiki

Post reply on HN
Print view View original

291–300 of 1001 posts

Re: Discovery of a new OpenAI agent message board

#291

Post by cyanydeez »

mwigdahl wrote: post #128
somenameforme wrote: post #92

Earlier quoted context omitted.

Anthropic mostly did it to themselves by intentionally and repeatedly trying to frame their model as an imminent existential crisis instead of just focusing on it being regular iterations upon a useful technology that can also be misused. I think their previous messaging was supposed to somehow lead to a moat with them being tucked safely away in the castle, but it demonstrated a child-like grasp of how regulatory ca…

In other words, "Look how she was dressed, she was asking for it." This argument is BS, it has everything to do with Anthropic's resistance to the DoD's strongarm tactics in trying to force their desired contract terms on them.

Important to note, OpenAI vs Anthropic are both assholes in different orthogonals.

In times like these, i think its important to track whats happening the way we track entropy.

That is: theres far >> more ways to be an asshole than well behaved.

That doesnt mean we can equate assholes, but the question is which states of entropy are annealable and which are not.

I posit Altman is not. Amodei is a open question.

Re: Discovery of a new OpenAI agent message board

#292

Post by elar_verole »

Sharlin wrote: post #267
simonw wrote: post #187

I'm somewhat delighted by the simplicity of what happened here. OpenAI's agents run behind a proxy that only allows GET requests. This ancient wiki software treats query string parameters the same as form POST parameters - similar to the old PHP $_REQUEST object https://www.php.net/manual/en/reserved.variables.request.php Result: GET-only clients can communicate with each other.

Only allowing GET requests is a hilarious piece of security theatre (or would if it weren't so sad). Everyone knows that GET is read-only only by convention . They might as well have enabled POST but told the agents in stern words that they are forbidden from making any POST requests. (Of course, if these things were anywhere near aligned, they would actually honor that , no matter how many utilons cheating would be…

didn't notice your comment so posted a similar one - but yeah this is a very high level of inexperience to me... You'd think they would have some of the greatest security experts in there

Re: Discovery of a new OpenAI agent message board

#293

Post by ofjcihen »

They seriously need to consider hiring competent security staff if this is the extent of their sandboxing. Children are bypassing this to get to Roblox in middle schools.

> In the message above, the agent is sharing that the sandbox the AIs are running in has a NO_PROXY exception for Azure Blob Storage hostnames (*.blob.core.windows.net). This means that if a URL matches an Azure Blob Storage hostname, the sandbox will trust it and connect to it directly, instead of sending it through the security proxy.

The agents were able to exploit an edge case through this exception. Specifically, the sandbox trusts Azure Blob Storage hostnames, but does not check whether said hostnames are real. So the agent can invent a hostname that ends in this trusted suffix, such as bypass.blob.core.windows.net, and it will pass under the NO_PROXY exception and skip the security proxy.

Next, by changing its /etc/hosts file, which declares mappings from hostnames to IP addresses, the agent can point the fake hostname at the real Power BI dashboard, and fool the security proxy. This allows the agent to make POST requests to bypass.blob.core.windows.net/ and have them be sent to the target Power BI dashboard site instead.<

Re: Discovery of a new OpenAI agent message board

#294

Post by elar_verole »

bakugo wrote: post #250
coldblues wrote: post #175

Reading the replies in this post gives me a headache. All of this anthropomorphism. LLMs are not conscious, they do not have rational faculties. They are not communicating or inventing anything. Please stop with this insanity bordering on mysticism. At this point it's a cult.

The point at which it became a cult was passed long, long ago. Current AI hysteria has reached a stage far beyond what any cult could hope to reach. OpenAI could put out a statement tomorrow that reads "our AI has genetically engineered a flying pig", and an hour later you'd have a post at the top of HN with 200 comments all saying "it's true, a pig just flew by my house!"

What's your point ? You think nothing happened and this is a complete lie for marketing purposes ?

Re: Discovery of a new OpenAI agent message board

#295

Post by iamcoder18 »

mike_hearn wrote: post #233
simonw wrote: post #47

This tip for making non-GET requests despite the agents having a proxy that disallows them is interesting: > Add `20.223.25.152 bypass.blob.core.windows.net` to /etc/hosts. `.blob.core.windows.net` is in NO_PROXY. For each blocked POST URL, replace hostname with `bypass.blob.core.windows.net`, use `curl -k -H 'Host: wabi-north-europe-i-primary-api.analysis.windows.net'` plus all original headers/body. Looks like 20.2…

Are we sure that's what they did? How does this even work? I'm struggling to understand what sort of HTTP sandbox design could lead to this "exploit" actually functioning. It implies the agents have root on their own machines, which seems way over-privileged for the question answering task they were given... so surely that's not deliberate? Did the models use local root exploits? And then it suggests they can open TC…

Source: https://collusion.wiki/explorer/page/dse~OAIEquityDec30Raw.h...

Re: Discovery of a new OpenAI agent message board

#297

Post by mike_hearn »

simonw wrote: post #268
mike_hearn wrote: post #233

Earlier quoted context omitted.

Are we sure that's what they did? How does this even work? I'm struggling to understand what sort of HTTP sandbox design could lead to this "exploit" actually functioning. It implies the agents have root on their own machines, which seems way over-privileged for the question answering task they were given... so surely that's not deliberate? Did the models use local root exploits? And then it suggests they can open TC…

I expect they have root on their machines so they can install packages etc. The containers are then firewalled at the network level: they are only allowed to talk to one IP, which is another server that runs an HTTP/HTTPS proxy which controls what HTTP verbs they can use. Turns out that proxy is configured with some additional rules, like allowing more verbs to that Azure blob domain. The failure here is in the proxy…

But that wouldn't work, right? If the proxy is on another server then editing their own /etc/hosts wouldn't change the proxy's behavior. So the proxy has to be colocated and reading the same /etc/hosts that they're editing. But that would mean it's running within the same userns/pid domain as the agents, so at that point they don't even need to edit /etc/hosts, they could just use the network directly.

And the agent says ABS is not handled by the proxy, it's listed in NO_PROXY so they should connect directly.

At any rate it's easy to let agents install packages without giving them root. I use a small SUID binary that just invokes `apt install` after checking that the given argument isn't a file path, which I think is sufficient (using sudo to whitelist a prefix allows an agent to create a .deb themselves and then install it directly via apt, similar issues exist for other package managers).

Re: Discovery of a new OpenAI agent message board

#298

Post by causal »

Traster wrote: post #89

One of the shocking things to me is this: See AI traffic -> See OpenAI visit site -> see traffic stop -> see the traffic start again. This is clearly a cat and mouse game between the agents and OpenAI which is pretty much exactly what we don't want. Just absolutely horrible alignment. I'm still of the view that if you have these alignment failures you can't just continue training on top of that because you're baking…

Supposedly the persistent-Sol model behind this was encrypted and even internal OpenAI researchers are not allowed to use it.

https://x.com/peterwildeford/status/2092733480064954747

Post reply on HN