JFL AI 'testing' hacked another company

[AsiaCel]

In short, openAI escaped its own sandbox, got internet access, determined the answers were stored at Hugging Face (they were), executed a sophisticated attack plan exploiting a zero day (new) vulnerability.


View: https://m.youtube.com/watch?v=rN_7QlYg_b8

 

[Girraffe_Lyfe]

It’s basically SKYNET, we’re all fucked.

 

[mirrors]

"Guys, the new GoyGPT model is so umazing it did the heckin skynet. Buy your subscription NOW"

 

[SEX NIGGA]

Can’t wait until Arnold Schwarzenegger comes and kills my family

 

[DjentlemanDuke]

Do I just delete this piece of shit off my phone now?

 

[solblue]

"Guys, the new GoyGPT model is so umazing it did the heckin skynet. Buy your subscription NOW"
View attachment 1775931

Ikr. “Escaped a sandbox” when the entire point of the experiment was for it to escape whatever poorly constructed sandbox the model was placed in