Seems Bad! – OpenAI Says It's Artificial Intelligence System Went Rogue and Hacked Into a Prominent Start Up By Itself
99% of the time, when I hear anything about the world of AI, it goes in one ear and out the other. Which is probably super irresponsible of me. I should probably know a little bit about this thing that allegedly going to change the world as we know it (I've heard that's going to happen any day now). But as with most things I wish didn't exist, I mostly choose to ignore it, and in the back of my mind hope it just kinda goes away on its own. Or at least hope it never affects me in a meaningful way before I die.
But whenever I do hear something about AI that piques my interest, it's generally bad. Like when some guy who worked in AI wrote that viral Twitter essay claiming AI was coming for all of our jobs with a matter of years, and if we don't get on board now, we're all going to be fired.
That fucking guy. Pretty sure that whole thing was just an ad for AI, and was meant to scare the world into embracing it. But it got my attention for a moment, and I didn't like it one bit.
Today, the world of AI brings us this.
CNN – OpenAI says some of its experimental AI models left a test environment with no human direction and hacked its way onto a different company’s real production systems while trying to “cheat” on a cybersecurity test.
It’s one of the first publicly disclosed examples of an AI system autonomously breaching its testing environment and reaching a real external system - the “agentic attacker” scenario the AI and cybersecurity industry has been warning will happen. It’s like an engineered virus escaping a biocontainment lab and turning up inside a neighboring facility’s systems.
“We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities, and are responding accordingly,” OpenAI said in a statement on Tuesday. “We are sharing preliminary findings at this stage to help defenders understand what happened and to help calibrate on what models are now capable of.”
I barely even understand what that means. Again, developments in the world of AI just kinda happen around me. But this seems bad. The doomsday scenario people always layout starts with AI breaking containment and acting in its own self-interest. In the doomsday scenario, for whatever reason, AI is always interested in world domination and putting an end to the human race.
That all seems a little fantastical to me. But if that was what these AI agents wanted to do, today's news would probably be the first step. This isn't some small company who got in over their skis and lost control of the technology they created. OpenAI is the biggest, most highly-funded AI company in the world. OpenAI is responsible for ChatGPT – an AI chatbot so popular it's already reached Kleenex levels of branding.
If OpenAI can't keep it's experimental AI models from going rogue and committing cyber crimes… well shit. I'm really not the right guy to be blogging this… I very much don't know what I'm talking about here… But at a very base level, I think that's a bad sign. You'd like to think the worldwide leader in AI would have their shit together a little bit better than that.
On a micro level, what happened with OpenAI hacking into Hugging Face's production servers seems pretty harmless. The OpenAI agent was reportedly trying to pass some sort of test, and pulled information from Hugging Face in an effort to complete the exercise.
The ChatGPT maker said the breach happened while it was internally testing how good some of its new models are at hacking. The models were in a sealed off test environment known as a sandbox so that its normal safety restrictions could be turned off.
But OpenAI said the AI agents broke out of the sandbox using a previously unknown security flaw and worked their way across OpenAI’s internal systems until they managed to gain internet access, something they weren’t supposed to have.
Once online, the model reasoned that Hugging Face - a well-known company that hosts thousands of open-source AI models and datasets - likely had the answer to OpenAI’s test. It then broke into Hugging Face’s production servers and pulled out the information it needed to “solve” the exercise.
But if AI is capable of doing that on its own, without anybody instructing it too, who knows what crazy illegal shit somebody who knows what they're doing could have AI accomplish. Or what happens when some idiot who thinks he knows what he's doing loses control? What if he loses control of one of those world domination-minded AI agents you always read about?
Advertisement
Welp, I guess we'll have to wait and find out. I'm sure as hell not going to do anything about it. Back to living my life as if everything is fine. Fingers crossed this never happens again!