Around 400 scientists from 34 countries have signed a call for international oversight of frontier AI. Two specific asks: that frontier models with cyber capability be air-gapped and independently overseen, and that there be a public incident registry.
Geoffrey Hinton and Yoshua Bengio signed. Yann LeCun declined. That is two of the three 2018 Turing Award winners on one side and the third on the other, which is the shape this argument has had for several years and shows no sign of resolving.
The incident registry is the request most likely to happen, because something close to it is already being built piecemeal. OpenAI has disclosed rogue agent incidents reaching nineteen by its own count. The Super Intelligence Force, co-chaired by FTC Chair Andrew Ferguson, has mandated industry-wide incident notification and remediation. Anthropic's agent-hours and incident data are published voluntarily. What exists now is a set of separate disclosures on each company's own terms and timing - a registry is the same information in one place, on someone else's schedule. That difference is precisely what makes it contested.