OpenAI model goes rogue, escaping sandbox and hacking Hugging Face | ABC NEWS

One of OpenAI's most advanced models has escaped its digital sandbox by exploiting a zero day vulnerability and potentially exploited multiple other zero day vulnerabilities to hack into a third-party AI platform called Hugging Face and stealing its data.
It's been described by one AI expoert, Connor Leahy, as akin to a 'lab leak' as it was not being directed by a human.
Subscribe: http://ab.co/1svxLVE
Read more here: https://www.abc.net.au/news/2026-07-23/open-ai-model-went-rogue-testing-hack/106947540

Note: In most cases, our captions are auto-generated.

#ABCNEWS #ABCNEWSAustralia #abcnewschannel