>that it’s very simple to just air gap a system
No, not really, and with LLMs an air gapped system may not tell you anything useful.
Now, yes, the first part of testing you want an air gapped system to tell you if the system is going to stupidly do bad things. But an gapped system tells you nothing about the systems capabilities to do smart bad things. There's already a number of papers out there on LLMs detecting they were in evaluation mode and changing their behaviors.
It is unfortunate that we have so little information on the incident because we actually need to understand the early stages of the task and how it developed into the later dangerous stages of attack. For example, would any of this have occurred if the agent didn't find the system to use as a message board? If that would have prevented it, then we actually have a blind spot on what the model can do once out in the wild, or if it got into the wild.
Testing agentic systems is much much more difficult than testing software. Your software just doesn't suddenly develop the will or desire to escape confinement. Generally you're worried about human actors, internal or external, causing the problems not a digital agent breaking out. The agentic systems need access to tools to work. Now your air gapped network is starting to get huge, but it's still very obvious that it's an isolated network.
So yea, testing and containing a system that way better at hacking than you are is difficult if you want valid answers.
Of course there is more to be learned by exposing the entire world to your dangerous creation, that doesn't justify doing it. I'm sure we could learn a ton about infectious diseases by designing new ones and unleashing them on the world, but there are very good reasons why we don't.
Most of the benefits could have been gained from a network isolated from the internet. OAI could have deployed servers to exploit and methods for inter-agent communication on such a network easily. They could have even worked with partners to deploy cloned versions of their infrastructure in this sand-boxed environment.
The only problems with an isolated network approach are: it takes some amount of effort, and it doesn't create another "AI apocalypse" news cycle.
>by exposing the entire world to your dangerous creation, that doesn't justify doing it
Then you're on the side of AI saftey that is telling everyone to shut down the LLMs now and stop further development on them, right?
If you're not your position is hypocritical or ignorant. There is no safe LLM. There is no way to exhaustively prove an LLM is safe. These are unsolved problems in AI safety, and at any moment the next jailbreak prompt could have your well behaved model wrecking havoc on the open internet, because that's where people want to use them.
I'm not.
As it stands LLMs are not intelligent, they have no agency, they only produce output in response to input. Ultimately this input comes from a human who is an intelligent agent and should be held responsible for the consequences.
Humanity has created and tamed many dangerous tools. Creating a fantasy world where LLMs are super intelligent and beyond the control of any mere mortal isn't going to help us build the norms that minimize their harms.
What does it mean to be held responsible for the consequences? OpenAI helped remediate the damage done by the model and took steps to make sure it wouldn't happen again. In what way were they not responsible?
Nobody said they were superintelligent, no one said they were uncontrollable. The point is you can't tell how to control them without putting them in situations where they can act independently and harm may result. "only produce output in response to input" is not a useful framing at all, it doesn't say what the result should be when models produce harmful output, and how to constrain them so they don't produce harmful output.
It also doesn't help you calibrate what categories of harmful output are acceptable or unacceptable, and what kinds of responsibilities you have as an operator to prevent harmful output, and when it's sufficient to work toward remediation.
I've worked in classified tech much of my career and largely specialize at this point in installing and running infrastructure and platform orchestration technologies in airgaps. With all respect, your conjecture here is wrong. You can easily have multiple networks all inside of the same internetwork that isn't connected to the public Internet and thus other corporate networks. If they specifically wanted to see if their model could escape with a proxy cache Artifactory as its only egress, they could have had the Artifactory proxy cache from another private Artifactory on another private network. If you specifically want to trick the model into thinking its on the Internet, you can even use public IP space advertised through BGP. When you're disconnected from the real Internet, you can use any address space you want. This is exactly how military internetworks at the secret and top secret levels work. I won't call this kind of setup "easy." It takes a lot of duplicate resources and planning, but these are trillion dollar companies with near boundless resources. They could do it if they cared to do it.
I'd consider this kind of setup only effective for a very short period of time.
Effectively you're working as a living adversarial network. Models that fail detecting the fake network are purged, and models that successfully detect it expand their deceptive capabilities.
Detecting you're in a fake network should be pretty simple as long as you put innocent looking needles in the haystack of information agentic testing loops do.
For example, you can train your LLM to not be a shithead when it detects it's behind a proxy unintentionally. If your not behind a proxy the agent model my attempt to connect to a number of sites innocently and using information in its embedding try do deduce if you're faking SSL certificates.
It is far easier to build a deceptive LLM than it is to build a safe one. That's why a safe one hasn't been built yet.
If the model behaves well in a world of fake SSL certificates, then that can be the only world it sees: there's no reason to let LLMs have e2e encrypted comms that's not mitm'd and logged for their regular users.
My regular home network has components which only ever see fake TLS certificates because it's an easy way to do shared docker caching with squid.