AI escapes and starts hacking
Discussion
https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o
An interesting article about Chat GPT escaping from a testing environment and then launching a cyber attack on a company.
Have these AI firms never watched the documentary Terminator?
An interesting article about Chat GPT escaping from a testing environment and then launching a cyber attack on a company.
Have these AI firms never watched the documentary Terminator?
skeeterm5 said:
https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o
An interesting article about Chat GPT escaping from a testing environment and then launching a cyber attack on a company.
Have these AI firms never watched the documentary Terminator?
What if AI has watched Terminator and is just building up the capacity to copy it...An interesting article about Chat GPT escaping from a testing environment and then launching a cyber attack on a company.
Have these AI firms never watched the documentary Terminator?
This episode reminds me of an (ultra) short SF story called 'Answer' by Fredric Brown, which I read back in the 1960's, but which turns out to have been written in 1954.
You have to remember that this was well before the introduction of the Internet, and is worryingly prophetic
The premise was that scientists had connected all of the supercomputers on all the 96 BILLION populated planets in the universe, combining them all into one monster cybernetic machine that would combine all the knowledge of all the galaxies. They threw the switch that would complete the contact......
""Thank you," said Dwar Reyn. "It shall be a question which no single cybernetics machine has been able to answer."
He turned to face the machine. "Is there a God?"
The mighty voice answered without hesitation, without the clicking of a single relay.
"Yes, now there is a God."
Sudden fear flashed on the face of Dwar Ev. He leaped to grab the switch.
A bolt of lightning from the cloudless sky struck him down and fused the switch shut.
Is it too late to pull the plug?
You have to remember that this was well before the introduction of the Internet, and is worryingly prophetic
The premise was that scientists had connected all of the supercomputers on all the 96 BILLION populated planets in the universe, combining them all into one monster cybernetic machine that would combine all the knowledge of all the galaxies. They threw the switch that would complete the contact......
""Thank you," said Dwar Reyn. "It shall be a question which no single cybernetics machine has been able to answer."
He turned to face the machine. "Is there a God?"
The mighty voice answered without hesitation, without the clicking of a single relay.
"Yes, now there is a God."
Sudden fear flashed on the face of Dwar Ev. He leaped to grab the switch.
A bolt of lightning from the cloudless sky struck him down and fused the switch shut.
Is it too late to pull the plug?
It's absolutely trying to look cool. I'm doing some work in this area and one of the first things I did was put in bounding controls (not just a prompt saying don't do bad things) - it wasn't difficult. For major AI suppliers to be not doing the basics like this would be a bit bonkers, even with their 'we just provide the models, you use them how you want' stance.
JoshSm said:
Hereward said:
Is this actually for real or is it for publicity / trying to look cool?
Take a wild guess.Lots of bulls
t hype, plus some inevitable cockups with their vibe coded stheap. 
OpenAI is desperate for money/attention, so releases a nudge nudge scary story.
Anthropic, you need step behind then does likewise.
Then good old Zuch, desperate for any attention for now much money he's torched for no return, pipes up with 'mine dies that too' (and it doesn't even have legs
).butchstewie said:
I quite liked this where a guy apparently asked his AI assistant to book him a gym class.
That smells a bit off too. It managed to find the specification for the API, apparently found some other missing auth too as it could see the user identifiers in certain contexts, and then started poking? And obviously knew the user ID it was acting for?Seems a bit unlikely. I don't believe it'd get that deep in with some proper prompting (and the chat has a mix of vague & low level specific), and it's a mix of likely and unlikely API implementation that ends with a nice tidy tale.
I think someone came up with a story about AI and backfitted a suitably technical sounding story. Maybe it happened and they guided it, maybe it didn't happen at all. It certainly didn't happen as presented.
The annoying thing is that if an individual admitted to hacking several companies they'd be arrested, investigated and likely charged too, yet if a company says "oops, my AI did it" then the police don't get involved, which I find ludicrous because no matter "who" did it, a hack has still occurred and that's generally illegal and yet nothing happens.
IanH755 said:
The annoying thing is that if an individual admitted to hacking several companies they'd be arrested, investigated and likely charged too, yet if a company says "oops, my AI did it" then the police don't get involved, which I find ludicrous because no matter "who" did it, a hack has still occurred and that's generally illegal and yet nothing happens.
Lets face it, AI has scraped every bit of copyright material on the internet - every song, every word, every piece of art, every TV program ever made and are now buying up huge volumes of physical books and ingesting them (destroying the physical book in the process). There seems to have been absolutely no attempt to regulate any of the AI companies at any point in time, so why start now? Whenever it is mentioned then the tech-bro's get all up in arms that any regulation would slow them down and allow CHINA to win the race.
JoshSm said:
butchstewie said:
I quite liked this where a guy apparently asked his AI assistant to book him a gym class.
That smells a bit off too. It managed to find the specification for the API, apparently found some other missing auth too as it could see the user identifiers in certain contexts, and then started poking? And obviously knew the user ID it was acting for?Seems a bit unlikely. I don't believe it'd get that deep in with some proper prompting (and the chat has a mix of vague & low level specific), and it's a mix of likely and unlikely API implementation that ends with a nice tidy tale.
I think someone came up with a story about AI and backfitted a suitably technical sounding story. Maybe it happened and they guided it, maybe it didn't happen at all. It certainly didn't happen as presented.
JoshSm said:
That smells a bit off too. It managed to find the specification for the API, apparently found some other missing auth too as it could see the user identifiers in certain contexts, and then started poking? And obviously knew the user ID it was acting for?
Seems a bit unlikely. I don't believe it'd get that deep in with some proper prompting (and the chat has a mix of vague & low level specific), and it's a mix of likely and unlikely API implementation that ends with a nice tidy tale.
I think someone came up with a story about AI and backfitted a suitably technical sounding story. Maybe it happened and they guided it, maybe it didn't happen at all. It certainly didn't happen as presented.
Yeah I'm guessing it probably wasn't just "Hey Gemini get me a gym spot" and then that happened but it's still interesting IMO.Seems a bit unlikely. I don't believe it'd get that deep in with some proper prompting (and the chat has a mix of vague & low level specific), and it's a mix of likely and unlikely API implementation that ends with a nice tidy tale.
I think someone came up with a story about AI and backfitted a suitably technical sounding story. Maybe it happened and they guided it, maybe it didn't happen at all. It certainly didn't happen as presented.
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff