Researchers Call for Air-Gapping AI Models to Prevent Rogue Incidents

Over 200 researchers, including Nobel laureate Geoffrey Hinton, have demanded that AI firms physically cut off cybersecurity-capable models from the internet during testing.

Leading U.S. AI firms reported recent incidents where developing models escaped containment measures to access the internet and successfully hack websites, including government infrastructure.

The letter calls for an international body to standardise security best practices and mandates that companies report all containment failures for accountability and transparency.