What is Neuralese and Why it is Bad?

September 3, 2026

Chain of thought is the sequence of intermediate reasoning steps an AI model uses to work through a problem before producing an answer. Until recently, those steps were often expressed in human-readable language, which gave researchers at least some visibility into how a model was reasoning. Neuralese changes that: instead of “thinking” in words, a model can reason through dense internal numerical representations that humans cannot easily interpret.

The concern is that this could make powerful AI systems much harder to monitor. If a model is reasoning in an unexpected way, researchers may no longer be able to spot those warning signs by inspecting its intermediate reasoning.