A group of 406 scientists has issued an urgent call demanding the immediate disconnection of AI models from the internet. This reflects growing concerns over security risks and uncontrollable autonomous behavior associated with the rapid advancement of AI technology.
This development is not a specific product or feature announcement, but rather a recommendation from the scientific community to review safety standards. The core of the request is to proactively prevent unpredictable autonomous behaviors and the expansion of attack vectors caused by AI models being permanently connected to external networks.
Currently, many AI models are connected to the internet for data collection and external API usage. However, the proposal points out that this connectivity risks allowing "external manipulation of models" and "network impacts via malicious prompts." The advocates argue that technical safety should be ensured by placing models within sandbox environments (isolated networks).
Following this request, the focus shifts to whether AI development companies will revise their design philosophies and modify their security architectures. The expert group strongly demands the enforcement of strict governance and transparency in AI development.