Fast takeoff

In plain English

Takeoff describes how quickly capabilities rise once systems approach human level. In a fast, or hard, takeoff a system crosses from roughly human ability to far beyond it in a short window, usually because it starts improving itself or because adding compute pays off far more than expected. There may be only one leading system when it happens, and the world may not notice until the jump is over.

The main argument for a fast takeoff is recursive self-improvement: research done by SI is faster than research done by people, and the gains compound. The main arguments against it are physical bottlenecks, such as chips, power, data centers and the need for real experiments. Forecasters and lab leaders disagree sharply about which effect wins, and many estimates moved earlier as reports of SI building SI grew in 2026.

Why it matters for rogue SI

A fast takeoff gives regulators, labs and the public almost no time to notice warning signs and respond. Most plans for containing a rogue system assume there is time to test, pause and correct; a fast takeoff removes it.