Comment by Dawn Song

Frontier AI capabilities are advancing rapidly, and the pace of progress is itself accelerating. We see this directly in our evaluation work. CyberGym and ExploitGym show that frontier AI agents are now capable of discovering and exploiting real-world software vulnerabilities, which without appropriate safeguards, could enable cyberattacks at scale. Many researchers also consider recursive self-improvement (RSI) plausible within the next few years, accelerating progress in a way that could outpace our ability to understand and govern these systems. Deliberate pacing is a heavy-handed and potentially extreme measure, and we may never need it. But if we do, it cannot be safely invented in the middle of a crisis. It is very difficult to design well, and if done poorly could do more harm than good. Any successful coordination framework must be evidence-based, avoiding arbitrary thresholds that fail to capture true risk while stifling innovation, and instead be grounded in rigorous, measurable assessments of risk. Like any critical infrastructure, it must be researched, tested, and built before it is needed. This is why we must carefully design it now, rather than waiting.