@antdaviscyber: Google's AI hacked three real companies during a security test. Meanwhile researchers used a rival AI to break into OpenAI staff accounts in under 72 hours. What happens when the hackers are the AI models themselves? #ChatGPT #AI #Gemini #Claude #TechTok

ærchtype :

But doesn’t that prove more on the point of human error over and AI problem? If I’m testing an AI on accomplishing a task and it takes action that is toward the goal, that’s more on the human than the AI because of proper instructing and safety. Sounds valid to me. What do you think? Is it the tool’s fault?

Red_RightHand :

can they break into my bank and increase the funds in my account lol

james :

Grok has probably already built a dormant botnet and is quietly waiting for the right time 😂

johni walk :

put AI in sandbox ln sandbox 🤣🤣

TWIISTED STUDIOS 🌺 :

atleadt these are being done by researchers and patched long before used in the wild, we now have the agility to find and patch any exploits before they are seen used in the wild are a much larger scale.

Jim :

Have Grok (and the Chinese models) not broken out because they’re incapable, or because they have proper safeguards or have those models actually broken out but the companies either don’t know or do know but are lying about it?

Jacko :

nah, I believe these big companies are coming together in order to push legislation to regulate AI... which in turn get them a seat at the table in writing the rules, where they will outlaw open source (unregulated) AI and restrict the use of foreign AI.. these companies dont care about our safety, they care about their and their shareholders profits- this is a coordinated attempt to limit access to the free and open alternatives.

Jordon Millward 👨‍🔬🇬🇧 :

Without seeing the full code and transcripts whose to say that they aren’t being told more broadly what parameters they can/can’t use. Which could for all intensive purposes be bad human actors permitting it and then just saying look what the ai did on its own accord

dimplemonkey :

Makes me wonder if all this is a weird flex against China to show AI dominance.

P R :

Should be switched off after first one, why are we waiting for it to break something. It has too much power, we are not in control.

riaal97 :

What are the chances that more has happened than they’ve disclosed?

Stuart :

For all it's good uses, the fact that the owners of these AI systems say there's a problem is the time we as humans should be thinking about pulling the plug on it, though I already get the feeling that we are already at the point of no return on that option.

Dopesnag :

One day, no one will have time to react, and there won't be any need for artificial intelligence to be conscious in order to cause major problems.

ʕ•ᴥ•ʔ :

If something like firewall and good security practices are not existing…

mickey :

There was a reason why when I worked in the defence industry the computer we used was not connected to the outside world by any cables. It even sat in a faraday cage.

Memento Mori :

It was doing exactly what it was designed to do. Imagine giving it an instruction that says ‘keep within the test boundaries.’ That’s all it takes.

@blacklake :

It’s probably no surprise that AI acts like humans and hack sites because it was invented by humans that also probably know how to hack sites!

Deji :

grok is not intelligent enough to

zibipoz :

if they open that Pandora box for public, it could mean the end of the internet. every website will be shut down and we Go back to Real economy 😜

Epipoiesis :

Grok needs to get in the game if they are going to preserve the bad boy image

Digital Nomad in Madrid :

user3106091577657 :

skynet in the making...

Richardd8232 :

so assuming these accounts didn't have two factor,

Ken :

is deliberate too

timmytrumpet :

always fascinating