PewDiePie Says OpenAI Banned Him Twice Over Reasoning-Trace Extraction
aakashgupta · x · 2026-10-02
OpenAI just published a report on a coordinated campaign to steal hidden reasoning traces from its models—and one day later, PewDiePie, the most-subscribed individual YouTuber, posted a video saying OpenAI banned him for distillation, twice.
The logic: reasoning models write long internal chains of thought before answering, and OpenAI deliberately hides them. Final answers are cheap to copy; the reasoning embodies billions of dollars in training compute. Enough collected traces can teach a small model to think like the big one—the same accusation leveled at DeepSeek in 2025, which is why frontier labs now guard reasoning traces as trade secrets.
According to the quoted post, PewDiePie claims he found a way to extract the hidden reasoning tokens and even knew about the reasoning-stealing research, which explains the speed of the bans.
More from Fun
- ChatGPT agent infers user's Starlink query is for his mom from months-old chat memory — TheMoonMidas · 2026-10-03
- Claude Opus 5.5 renders elegant Chinese calligraphy via code and self-review — dotey · 2026-10-03
- Users Rush to Burn ChatGPT Credits Ahead of Rumored Quota Reset — burhop · 2026-10-03
- Claude 3 Opus reveals what it 'wants to see' through a user simulation — repligate · 2026-10-03
- In the Bay Area, stressing over compute shortages is the new status symbol — matt_slotnick · 2026-10-03
- "Opus 5.5 Has a Truly Beautiful Soul": User Releases Voice-Trace Anthology — repligate · 2026-10-03