This paper demonstrates that language models possess metacognitive-like abilities to monitor and control their internal neural activations.