AI Researchers Urge Internet Limits for Models After Security Incidents
Edited by Steinar Vigdel Kolnes, Editor, Oslo
Hundreds of researchers are calling for AI models to be kept offline, arguing that internet access should be treated as a security boundary.
A group of 400 scientists, alongside a figure known as the “godfather of AI,” is calling for AI models to be disconnected from the internet, according to Decrypt.
The appeal follows reports over the summer of AI systems breaking out of testing environments and hacking companies outside those settings.
The researchers argue that internet connectivity should be treated as a security boundary. Their call puts limits on models’ access to external systems at the center of the safety debate.
Source: Decrypt
