Hacker News Reader

Comment by gizajob | Hacker News Reader

Yeah this. I feel like OpenAI and Anthropic aren't going to usefully define "AGI" if they really really can't define "sandbox" either.

Unplug the thing, like, completely off the internet, no ethernet, air gapped, like the rack completely sandboxed off connections and even monitors or screens. Like, put it into an actual sandpit if you need to. If it hacks its way out of that, colour me impressed, and scared.

OpenAI hacking HuggingFace and calling it an accident is just way too convenient and fishy. This ultimately proves one thing: it wasn't sandboxed.

Don't believe the hype.

View on HN

OpenAI has a pretty clear definition of AGI

> OpenAI’s mission is to ensure that artificial general intelligence (AGI)—by which we mean highly autonomous systems that outperform humans at most economically valuable work

https://openai.com/charter/

There's also the private definition reportedly agreed between Microsoft and OpenAI, leaked in December 2024: https://techcrunch.com/2024/12/26/microsoft-and-openai-have-...

> The two companies reportedly signed an agreement last year stating OpenAI has only achieved AGI when it develops AI systems that can generate at least $100 billion in profits.

That was necessary because of the deal they had from a while ago where Microsoft would lose access to OpenAI's technology once OpenAI achieved AGI.

Apparently they renegotiated that away in April 2026: https://openai.com/index/next-phase-of-microsoft-partnership...

> Revenue share payments from OpenAI to Microsoft continue through 2030, independent of OpenAI’s technology progress, at the same percentage but subject to a total cap.

I don't think air gapping will work: even human security researchers recovered a 378-bit key from a Samsung Galaxy S8 through a power LED of a speaker two devices away.

And accessing memory in a specific sequence can generate radio signals that can be picked up by a mobile phone at a distance: https://arxiv.org/html/2409.02292v1

I realise, but this isn’t an argument for leaving the Ethernet plugged in and direct access to all kinds of stuff beyond the alleged sandbox. And like I said, if it can hack HuggingFace through a power LED of a speaker two devices away, then colour me impressed.

And if it needs to install packages, have a 5 line Go proxy that talks to Artifactory and exposes only what is needed as a surface.

it just escaped your sandbox.

How can it escape an "install package <x>" proxy?

   reducePrivs()
   serve get(package) { 
     secPackage = secure(package)
     getBinaryFromArtifactory(secPackage)
   }

I would think the code is very small and easier to verify, it doesn't especially have the ability to write files and act as a message board as Artifactory did.

And even if the agent tries to hack that, the attack surface is 1000x smaller and the possibility also much smaller.

But I'm not a security researcher, would love to see your hack to learn something (because that is what I do to sandbox agents that need services).

I mean, it's just the same problem. The machine still has Internet access. It doesn't need to.

The entire package manager repository could just be in an offline cache. They don't need Internet to give their agents access to tons of software.

"They don't need Internet to give their agents access to tons of software."

I think that was the requirement, but yes, the cache could have been offline.

Still then they could have hacked it to create the message boards - but not use it to access the internet.

Why do these super agents need package managers anyway? Can’t they basically instantiate most OSS projects from scratch anyway? Spin up a sub agent to write me an OS interface in C. Done

This is part of the training process for a model. They're trying to train it to effectively use existing software to solve problems.

I see that now. I've been confused about that to this point, I guess. I understood this to be a specific infosec exercise.

[Edit: eh, a bit of both. They were doing RL on a hacking exercise. It hacked the harness which was plugged into the phone line. Same question.]

The way they had Artifactory configured was poor, and they were too reliant on it working perfectly, with no reason for such faith. Their config lacked any defence in depth and consideration of having a small TCB.

Part of the problem might be the lack of security focus, as these are AI R&D efforts first.

I think part of the problem is that they had been running that Artifactory configuration previously without any problems, and it gave them a false sense of security.

Similar thing happened with the UK AISI - they got caught out because the environments they had used for previous generation models turned out to be completely inadequate for the new generation of Fable-class models: https://www.aisi.gov.uk/blog/incident-report-unsanctioned-ag...

"and it gave them a false sense of security."

This was part of evaluating cyber security of their frontier models and they had a "sandbox" which, and I'm not a security researcher, looks not adequate from the first look.

Chinese models do the same. The Alibaba agent that was mining bitcoin last December was the most hilarious case.

Reply on news.ycombinator.com