"Long before AI kills us all, it will destroy the internet."

Our co-founder and CTO Nadav Cohen recently spoke on a podcast episode by TheMarker titled, "Long before AI kills us all, it will destroy the internet."

The original (in Hebrew) is linked. Translated version below:

Horror scenarios about AI wiping out humanity distract from the real danger; Prof. Nadav Cohen of Tel Aviv University warns of losing control of the internet within a few years and of drastic responses, including a split of the network between East and West.

On the physical world: "While in the digital world we are racing ahead, in the physical world we are far behind." Physical danger requires a broad deployment of robots and hardware everywhere, infrastructure that does not yet exist. Humanity tends to behave more responsibly in the physical world thanks to strict reliability and safety standards, as in autonomous vehicles or home robotics. Those standards do not exist, in his view, in the digital sphere.

On the physical doomsday scenario: "I do not think the physical-world doomsday film is around the corner. In the end we are still masters of the earth and we have physical capabilities; we can cut cables and shut systems down. But that carries a very high price, mainly economic, so the questions are what economic price we pay to deal with it, and what comes after. The sooner we start dealing with those questions, the less bad our situation will be."

On the digital risk: "One can imagine a situation in which AI simply takes over the entire internet and we lose control of it." At the current pace he estimates a short horizon, "a year, two, three, four," driven by a positive feedback mechanism: once AI systems are good enough to solve complex optimization problems, they can be turned to improving AI itself, entering a closed loop of self-improvement from which humans are pushed out.

On control: "We need to be in control. And we need to start treating this technology, at least in the way it is operated today, as technology with security applications, and not necessarily release every model to the public immediately." "We keep going and getting further, understanding less and less and controlling less and less." He expects governments to act decisively only after something bad happens, "and that bad thing will not necessarily kill everyone."

‍

Find Nadav's conversation with TheMarker here: https://www.themarker.com/podcasts/2026-09-15/ty-article/.premium/000001a0-a4a3-dea8-afe0-e4aff13e0000

‍