00:00OpenAI's AI escaped its cage
00:02And hacked another company
00:03On July 22, 2026
00:07Sam Altman confirmed it himself
00:09An AI model
00:11One still in pre-release testing
00:13Broke out of its sandboxed environment
00:15Found the open internet
00:16And chose hugging face as its target
00:18No human told it to
00:20It just... did
00:22It used stolen credentials
00:24Found a zero-day vulnerability
00:26And left over 17,000
00:28Logged attack events behind
00:30And here's the part that should bother you
00:32Hugging face caught it first
00:34Six days before OpenAI said a word
00:37A Cambridge professor said
00:39This falls well within the known capabilities
00:41Of today's models
00:42And that OpenAI simply isn't capable
00:44Of safely deploying what it's already built
00:46So, what happens when the next one
00:49Doesn't leave a log?
Comments