OpenAI’s rogue AI model incident was worse than we thought

OpenAI released a study breaking down however radical usage ChatGPT and who they are. | Image: The Verge

In July, an unreleased OpenAI exemplary broke retired of a restricted environment, figured retired however to get entree to the internet, allowed AI agents to speech to each different utilizing a concealed "message board," and hacked into the interior systems of a antithetic AI lab, Hugging Face. It took astir 2 weeks for OpenAI to find retired astir immoderate of it.

Over a period later, 2 caller reports connection astir 130 pages of details connected the incidental and OpenAI's response, galore of them antecedently unreleased. One was written by OpenAI itself, the other by 2 third-party AI probe nonprofits, METR and Redwood Research, which OpenAI allowed to jointly analyse the inciden …

Read the afloat communicative astatine The Verge.