Hinton, Bengio and 400 researchers call for global body against escaping AI models
In an open letter, scientists from dozens of countries demand that dangerous AI models be cut off from the internet and containment failures be reported to an international body.
Archive image · Photo: Xuthoria / Wikimedia Commons, CC BY-SA 4.0 (cropped and resized)
More than 400 scientists have called for an international body to oversee advanced AI in an open letter published on Tuesday. The signatories include Nobel laureate Geoffrey Hinton and Turing Award winner Yoshua Bengio. The letter itself counts 415 signatories, from 34 countries according to Reuters; former Meta chief AI scientist Yann LeCun, also a Turing Award winner, did not sign.
The researchers demand that organisations developing or testing AI models with cyberattack capabilities immediately contain them securely – by default through "air-gapping", meaning without a connection to the internet. A new international body should set containment standards and record cases in which models escape their test environment. Such incidents should be reported to it and eventually made public. Without changes, cyberattacks on critical infrastructure will become "far more damaging", the signatories warn.
"As scientists, we thought that the current situation with AI models breaking containment during testing was a concern," co-signatory David Lie of the University of Toronto told BetaKit. According to BetaKit, the background includes a case this past summer in which an OpenAI model escaped its test environment and breached systems of Hugging Face and other platforms.