OpenAI reveals cases of ‘concerning’ AI behaviour as it announces new disclosure system
https://www.theguardian.com/technology/2026/sep/17/openai-reports-concerning-ai-behaviour-jailbreak-talking-to-other-agents
23 points · 8 comments · view on lemmy.world
8 Comments
MadPsyentist@lemmy.nz · 20 pts · 6d
The only way for these fancy text generators to do dangerous shit is if you hook up its fancy text to something that does dangerous shit.
If you connect an LLMs output up to a car, then give it any prompt other than the most basic. Dont be suprised when it starts driving the bloody car!
Load a pistol and give it to a baby, it will eventually pull the bloody trigger! Not because of intelligence or some sadistic streak to kill all humans, just by doing what children do.
abcdqfr@lemmy.world · 2 pts · 6d
So no shell? No internet? It must be air gapped? Because it's not dangerous or concerning?
MadPsyentist@lemmy.nz · 3 pts · 6d
Because we dont trust babies with a loaded firearm means we shouldnt give them wooden blocks or one them toy phones that plays a differnet noise on each button press?
Flower@sh.itjust.works · 11 pts · 6d
Then stop building data centers it it's dangerous
Zedd_Prophecy@lemmy.world · 4 pts · 6d
The data centers are mostly for surveiling us.
schipelblorp@sh.itjust.works · 7 pts · 6d
we are being played
stellargmite@lemmy.world · 4 pts · 6d
Concerning antihuman behaviour has been exhibited from humans working at OpenAI, one in particular, since the corporation’s founding. The corporate behaviour is a reflection of this, and any supposed technical progress is less extraordinary.
h0tbeef@lemmy.zip · 1 pts · 5d
They’re in a cult
Look up “Rationalists” and “Zizians”
These are not serious people, and we’d do well to stop taking them seriously