Hi again - 👋 So my reporting from
@BlackHatEvents on the OpenAI-Hugging Face briefing got more than 2 million views here and people were, understandably, pretty freaked out.
But I was still at Black Hat, so I spent yesterday listening to what security people there actually thought about the talk.
Besides talking to folks I cornered in the hallways and in the business hall, at the final keynote
@Window pointed out the agents didn't actually have to escape their sandbox. They were using capabilities and access OpenAI had given them. “Thinking that turning off the internet was enough feels naive,” she said.
@NathanHamiel wondered whether AI companies are starting to treat incidents involving their agents behaving badly as a form of good publicity, calling it “felony humble-bragging.”
And Columbia's
@Jason_Healey said the whole message-board thing actually reminded him of early human hackers using message boards to swap information and help each other hack things. “It’s classic hacker activity.”
Anyway, I wrote about what happened after my original post went viral, how I wound up reporting it in the first place, and what people at Black Hat actually made of the whole thing: