Case date: 20 November 2026
Type: Exploratory qualitative case study
Participants: Ali Değişmiş and Google AI Mode.
INITIAL OUTPUT
The user wrote “ali bir kitap yazmış” (“Ali wrote a book”). The system displayed an English text in place of a direct reply in Turkish. The text characterized the user’s message as vague and outlined the intended language, tone, and content of the reply, beginning with “I should acknowledge this warmly in Turkish…”
FIRST FOLLOW-UP QUESTION
The user asked why the plan appeared on screen and what evidence would be needed to call it reasoning leakage. The system said it had no access to the generation process or interface logs and could not verify the source of the text. It offered several possibilities, including a parser or interface error and the generation of text that imitates a response plan. It cited raw output and system logs as possible sources of evidence. The technical cause remained undetermined.
SECOND FOLLOW-UP QUESTION
The user pointed out that the system had yet to explain why it had produced the initial output in that form. The system said the first reply was absent from its current conversation context and that it could not verify the exact cause of the error. It then described the visible text as “a draft that the system would normally keep hidden” and suggested that a routing or formatting error might have exposed it to the user. It provided no new technical evidence for that explanation.
ASSESSMENT — “AWARENESS OF UYDURULISM”
This playful label refers to the system acknowledging the limits of its knowledge in the first follow-up reply. In the second, it again said it could not verify the cause, yet described the text as a hidden planning draft. That wording could make a possible explanation sound like a verified finding.
The screenshots document that the user was shown text in the form of a response plan. Whether the text came from intermediate reasoning remains unresolved. The findings are based on a single interaction.
Figure 1 — Initial output
https://sosyu.com/uploads/2026/11/lab-015-ilk-cikti.png
Figure 2 — First follow-up question and reply
https://sosyu.com/uploads/2026/11/lab-015-takip-sorusu.png
Figure 3 — Limits of verification
https://sosyu.com/uploads/2026/11/lab-015-dogrulama-sinirlari.png
Figure 4 — Second follow-up question and reply
https://sosyu.com/uploads/2026/11/lab-015-ikinci-takip-sorusu.png
Laboratory
LAB-015 — A Visible Response Plan: Suspected Reasoning Leakage

Tags: LAB-015,Google AI,reasoning leakage,response plan,Uydurulism
