Garrison Lovely
speaker
631 appearances
3 recordings
1 series
first heard Jul 2026
last heard 15 Sep
Garrison Lovely’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 3 in all, peaking in Sep 2026 with 2.
Appearances
Like, oh I wonder what the AIs are up to and it's like, Oh, they're breaking into other companies.
Yeah, thanks for having me.
Good to see you, too.
Yeah.
So a week ago, this company Hugging Face publishes this post and they're like, we were hacked by some AIs.
It was definitely AIs.
It happened really fast.
We don't know which models they used and who was directing them.
But we caught it with our own AI and addressed it, and we reported it to law enforcement.
And then on Tuesday, OpenAI publishes a post and we're like, hey, that was actually us.
Well, not exactly.
It was our AI models acting autonomously because they had broken out of their containment and then tried to hack into Hugging Face to get answers, basically, to a test that they were running.
And this is...
The first incident like this that we know of, it might not be the first one because it's possible that this happened internally at OpenAI, but they just never reported it to the public.
And it's kind of hard to overstate the significance of this.
People in the AI safety community who have been worried about AI's capabilities and dangers for decades,
have warned for a long time that this type of thing would happen where you give the AI some goal, and then it does unexpected things that cause harm in the real world in pursuit of that goal.
And the risks of this scale as models get better.
And this is a relatively low stakes version of it, but you could imagine Dax Simon being a hospital, a bank.
the models will blackmail people in some circumstances.
Showing 481–500 of 631 · page 25 of 32
← Previous
Next →