In an interview with Sifted, safety expert Denis Shilov described the current push to build ever more powerful AI as 'the modern nuclear arms race'. His comparison is meant to capture the dynamic in which competing labs and nations feel compelled to forge ahead, even as the potential consequences become harder to predict.
Shilov argues that this dynamic is dangerous. Just as nuclear weapons required arms-control agreements and careful safeguards, he believes advanced AI needs deliberate limits and time for safety research to catch up. The alternative, he suggests, is a race that prioritises speed over caution.
The interview adds to a broader debate about whether the AI industry can regulate itself or needs external pressure to slow down. Shilov's framing suggests that the problem is not any single model, but the structure of the competition itself.