With Gemini v3.1 Pro AI Studio web interface, model thoughts leak “to the answer” the moment it “writes” $[0,\infty)$

Google AI Studio
1

Everything in the model’s thoughts after the leading $ of $[0,\infty]$ “is not hidden-by-default” anymore (hence — most probably — affects future rounds of the chat).

This happened in several recent dialogues (do not ask me why “it thinks about” $[0,∞]$ ! ;-]). One example is

https://aistudio.google.com/prompts/1kVL-fpZNKyF-mEzd29Vz1VDnPw1A9YTK

(This chat is not shared; should I?)

4

Correction: the actual triggering sequence is $[0,\infty)$ (the closing paren is ) — and not ] as initially reported).