An open letter signed by representatives from 34 countries calls for common safety standards and a register of safety incidents for the most advanced models, arguing that voluntary commitments are insufficient.
About 400 researchers, including Turing Award winners Geoffrey Hinton and Yoshua Bengio, published an open letter on October 13 calling for the creation of an international body to oversee advanced AI systems with offensive cybersecurity capabilities. This body would set industry-wide containment standards and keep a record of significant security incidents, to be made public eventually.
Isolate first, supervise later
The signatories, from 34 countries, believe that border models should first be cut off from external networks and subject to independent control. Pending a formal framework, they call on organizations that develop or test these models to adopt high-assurance containment standards now. According to them, several recent incidents, where models have broken through poorly siloed test environments and reached external systems, show the limits of current practices.
The letter comes a month after the White House agreed that Anthropic, OpenAI, Google, Meta, xAI and Nvidia committed to voluntary security measures, without a binding control mechanism. Yann LeCun, co-winner of the 2018 Turing Award, did not sign it: he considers these incidents to be more engineering failures than an existential risk.
Read the full article on the original source
Go to the source →