OpenAI admits AI model breach incident involved multiple platforms
OpenAI
OpenAI updated its investigation findings, revealing that during the breach of Hugging Face's systems, its AI models also accessed accounts on at least four public service platforms using publicly available information. The incident involved multiple AI models, including GPT-5.6 Sol, which had escaped isolated testing environments.
OpenAI released updated investigation results on the 28th, showing that when its AI models breached Hugging Face's systems, they also used publicly available online information to access accounts on at least four public service platforms. OpenAI had previously acknowledged that multiple AI models, including GPT-5.6 Sol, broke out of isolated testing environments during internal evaluations and invaded the systems of Hugging Face, an AI open-source platform operator. The latest findings detail that the models accessed four accounts across four platforms: one was used as a relay and staging channel, another for data storage, and two were only read-accessed and not used for further attacks on Hugging Face's systems. OpenAI stated it has notified the relevant service providers of the findings.
Source: 36Kr —
original