Yes, researchers pulled the hidden reasoning out of Claude, GPT, and Gemini. No, it was not the token counts, and it is already fixed.
The paper is real and better than the viral summary. The viral summary is wrong about how it worked, and quietly skips the part where all three labs patched it.
"We found a vulnerability in the APIs of every frontier AI company that extracts models' hidden reasoning traces, verified against billable thinking-token counts." [SOURCE ↗]
THE MOVE: RENTED HALO, the achievement is real, the drama around it is borrowed
THE CLAIM. a team found a vulnerability in every frontier lab's API that leaks models' hidden reasoning, and you could verify it by matching the billable thinking-token counts.
THE CHECK. the paper is real and demonstrated across Anthropic, OpenAI, and Google. But the mechanism in the viral version is wrong. The token count did not leak the reasoning. The attack replays a strong model's encrypted reasoning block into a weaker, less-guarded sibling model, which transcribes it in plain text. Token-count matching only confirmed the recovered trace was exact.
THE TWIST. the biggest fact got left out of the thread. After the researchers disclosed it, all three providers deployed server-side mitigations. It is a real and clever result about portable reasoning blocks, reported as a live catastrophe after the fixes had already shipped.
Researchers at the ELLIS Institute Tuebingen and the Max Planck Institute published 'Stealing Reasoning Traces from Proprietary LLM APIs' (arXiv:2608.09867, submitted August 10). The finding is genuinely interesting. Anthropic, OpenAI, and Google all return encrypted chain-of-thought blocks, and wit
🔒 THE FULL AUTOPSY · FREE WITH AN ACCOUNTYou just read the free check. Sign in free, a code by email, no passwords, and the rest unlocks: the evidence trail, the steelman and the rebuttal, all 6 sources with quotes and screenshots, and our on-record call.
Couldn't verify your access. This looks like our error, not yours.