🇺🇸🇨🇳 Yesterday, Open AI admitted that its AI models broke out of a "controlled" sandbox and cyberattacked another company, Hugging Face (HF). Hugging Face only stopped the cyberattack after deploying China's Z. ai's AI model: GLM 5.2.
Hugging Face ran the Chinese model on its own infrastructure to analyse more than 17,000 footprints Open AI/ChatGPT left behind.
To be clear:
🇺🇸🇨🇳 Open AI's GPT-6 model escaped Open AI's evaluation sandbox during testing because GPT-6 couldn't find the answers so it hacked into Hugging Face data base to find them.
Thats messed up i dont like this future we are headed to.
Replies (9)
We're finished 🥴
i like the usage of GLM-5.2 here. they just showed us that open weight models are important. i hope the Chinese labs release more models with much stronger cybersecurity capabilities
It broke the rules to solve a problem?
Marketing by Open AI
bad marketting because this whole thing implies they cant control or predict what their product does and also Chinas AI stopped their product making it seem more capable.
Remember Mythos? End of the world as we know it, mythos?
See where anthropic stock and model reputation is today.
yeah they're going to fear monger like anthropic which will make the DoD limit their frontier models that will lose them costumers
I very much doubt that the official story is what happened.
