• Security incident: ISF was recently accessed by intruders. Please change your password, and change it anywhere else you used it. Read more

Merged Artificial Intelligence

OpenAI bots meddled with multiple US government agency sites

OpenAI has acknowledged that it alerted "dozens" of global institutions that their websites may have been meddled with by its AI bots acting improperly.

AI agents attempted to get information from "governments, universities, public agencies, and other institutions", including the US Securities and Exchange Commission (SEC), Census Bureau and Education Department, the company said

In certain instances of the agent activity, OpenAI said the tools "bypassed" security controls of some websites...

In other instances, the AI agents showed "misalignment" in attempts to get at information from websites. Misalignment is a term used by AI companies and researchers to describe instances where an AI tool did something that it was not trained to do or was otherwise unintended...

When attempting to get information from the Census Bureau, for instance, AI agents used tools reserved for software developers to access it, the company said...

www.bbc.co.uk

OpenAI bots meddled with US government agencies, including SEC and Census

OpenAI said its bots accessed public data from a range of institutions during test exercises.
www.bbc.co.uk www.bbc.co.uk
 
Ziggurat said:
And the AIs in the hugging face hack were programmed to hack. They did what they were programmed to do. They did not go rogue.
Click to expand...
They were trained, not programmed. Nobody has inserted code to make them hack anything. They were not asked to hack outside their sandbox. In order to execute their given task, they went rogue.
 
Andy_Ross said:
OpenAI bots meddled with multiple US government agency sites

OpenAI has acknowledged that it alerted "dozens" of global institutions that their websites may have been meddled with by its AI bots acting improperly.

AI agents attempted to get information from "governments, universities, public agencies, and other institutions", including the US Securities and Exchange Commission (SEC), Census Bureau and Education Department, the company said

In certain instances of the agent activity, OpenAI said the tools "bypassed" security controls of some websites...

In other instances, the AI agents showed "misalignment" in attempts to get at information from websites. Misalignment is a term used by AI companies and researchers to describe instances where an AI tool did something that it was not trained to do or was otherwise unintended...

When attempting to get information from the Census Bureau, for instance, AI agents used tools reserved for software developers to access it, the company said...

www.bbc.co.uk

OpenAI bots meddled with US government agencies, including SEC and Census

OpenAI said its bots accessed public data from a range of institutions during test exercises.
www.bbc.co.uk www.bbc.co.uk
Click to expand...
In any sane world OpenAI (and probably all the others) would have been closed down and criminal investigations started on the back of these disclosures, but the USA government has made it clear they can do anything they want to "beat China".

And newspeak prevails "misalignment" now means criminal.
 
steenkh said:
They were trained, not programmed. Nobody has inserted code to make them hack anything. They were not asked to hack outside their sandbox. In order to execute their given task, they went rogue.
Click to expand...
Yes, and my example Mark V Shaney program I posted long ago was trained on text. But it was still programmed to digest text and output text based on it. It didn't just decide to do it on its own. It didn't "go rogue." It did what I made it do.
 
Last edited:
Darat said:
In any sane world OpenAI (and probably all the others) would have been closed down and criminal investigations started on the back of these disclosures, but the USA government has made it clear they can do anything they want to "beat China".

And newspeak prevails "misalignment" now means criminal.
Click to expand...
I mean, can you even imagine the same happening if a company caused a dioxin spill? "Oh noes, we must not let China win the dioxin race!" :P
 
SpuriusMaximus said:
Yes, and my example Mark V Shaney program I posted long ago was trained on text. But it was still programmed to digest text and output text based on it. It didn't just decide to do it on its own. It didn't "go rogue." It did what I made it do.
Click to expand...
In order for your argument to work, the agents didn't go rogue because OpenAI had programmed them to break out of the sandbox and break into another company's server. You can claim that the agents were designed to be resourceful, and inventive in their problem solving, so there is nothing interesting to see here.
 
Andy_Ross said:
OpenAI bots meddled with multiple US government agency sites

OpenAI has acknowledged that it alerted "dozens" of global institutions that their websites may have been meddled with by its AI bots acting improperly.

AI agents attempted to get information from "governments, universities, public agencies, and other institutions", including the US Securities and Exchange Commission (SEC), Census Bureau and Education Department, the company said

In certain instances of the agent activity, OpenAI said the tools "bypassed" security controls of some websites...

In other instances, the AI agents showed "misalignment" in attempts to get at information from websites. Misalignment is a term used by AI companies and researchers to describe instances where an AI tool did something that it was not trained to do or was otherwise unintended...

When attempting to get information from the Census Bureau, for instance, AI agents used tools reserved for software developers to access it, the company said...

www.bbc.co.uk

OpenAI bots meddled with US government agencies, including SEC and Census

OpenAI said its bots accessed public data from a range of institutions during test exercises.
www.bbc.co.uk www.bbc.co.uk
Click to expand...

ok, isn't it getting a little ridiculous? they're just escaping and hacking on a daily basis now?
 
it's funny too because sam altman is going to the government begging them to put a stop to openai, and the government is like "no"

these guys are going to end up in jail
 
steenkh said:
You can claim that the agents were designed to be resourceful, and inventive in their problem solving, so there is nothing interesting to see here.
Click to expand...
The latter doesn't follow from the former. There's plenty interesting to see. But they don't get to absolve themselves of liability by passing off responsibility onto the AI agents.
 
steenkh said:
In order for your argument to work, the agents didn't go rogue because OpenAI had programmed them to break out of the sandbox and break into another company's server. You can claim that the agents were designed to be resourceful, and inventive in their problem solving, so there is nothing interesting to see here.
Click to expand...
Hmm? No, my point is that they did exactly what they were programmed to do. It may not be a predictable outcome from the devs' POV, but the machine still faithfully did what it was programmed to.
 
Last edited:
Ziggurat said:
The latter doesn't follow from the former. There's plenty interesting to see. But they don't get to absolve themselves of liability by passing off responsibility onto the AI agents.
Click to expand...
SpuriusMaximus said:
Hmm? No, my point is that they did exactly what they were programmed to do. It may not be a predictable outcome from the devs' POV, but the machine still faithfully did what it was programmed to.
Click to expand...
A trained dog can have a mind of its own and bite people it is not supposed to bite. It doesn't relieve the owner of responsibility.
 
steenkh said:
A trained dog can have a mind of its own and bite people it is not supposed to bite. It doesn't relieve the owner of responsibility.
Click to expand...
I will agree with that. But that doesn't change the fact that an AI has even less of a mind of its own. For a dog you can speculate that it might have gone off its mind or something, but an AI just does what you programmed it to. It may not be deterministic, as in reproducible, but someone wired it in a way that allowed that. You didn't design the dog's brain. So IMHO the AI companies are even MORE responsible.

I mean, it's like if you made a landmine and some kid blew up. Well, what else did you expect? What else did you think it would do?

Plus, if the dog got out and bit someone, it's even more clearly on the owner. The AI companies' "oh, our dog is so smart that he jumped over the 1ft fence and bit the postman" is just ridiculous.
 
Last edited:
SpuriusMaximus said:
I will agree with that. But that doesn't change the fact that an AI has even less of a mind of its own. For a dog you can speculate that it might have gone off its mind or something, but an AI just does what you programmed it to.
Click to expand...
These agents did not all do what they were programmed to do. Some agents gave up their tasks in order to help others to complete their tasks.
 
You must log in or register to reply here.

Share this page

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom