The trick: Rented Halo
Yes, researchers pulled the hidden reasoning out of Claude, GPT, and Gemini.
No, it was not the token counts, and it is already fixed.
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.
Before you read on. Your call?
TRUE, BUT
3
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.
There’s more to this story.
Membership opens the full investigation, the strongest counterargument and what to do with what you’ve learned.
30 days free, then A$89 a year. Payment card required. Cancel before the trial ends to avoid the first charge.
Review my 30-day free trial →- At sign-up: 30 days free. Payment card required. One introductory trial per customer.
- Your trial ends 30 days after you start. Unless you cancel before then in Account → Manage subscription, we charge A$89 for the first year. It renews automatically at A$89/yr until cancelled.
- You can ask for a full refund within 14 days after any annual payment, renewals included, with no reason needed, by emailing hello@bskiller.com. This voluntary refund does not limit your rights under the Australian Consumer Law.
By starting your trial, you agree to the Terms.
BS Killer is published by Inferno Tech Pty Ltd, ABN 27 647 413 474.
Get tomorrow’s check free by email.
Couldn't check your access. That's on us.
The trick has a name
We call it Rented Halo: the achievement is real, the drama around it is borrowed. You'll see it again. Learn to spot it →
Receipts
- Supports arxiv.org:
we demonstrate across Anthropic, OpenAI, and Google
- Context arxiv.org:
By injecting an encrypted reasoning trace from a given model into a weaker, and less safeguarded model from the same provider
- Context arxiv.org:
By decoding 315,320 reasoning blocks scraped from public repositories
- Refutes cybersecuritynews.com:
deployed server-side mitigations
- Refutes simonwillison.net:
All model providers acknowledged the receipt of our report
- Context news.ycombinator.com:
nothing scientific here, they basically just figured out some real issues caused by bad engineering practice.
Open the Receipts Pack → What each source proves, every figure traced, and what would change our verdict.
