Start free trial

TRUE, BUTRented Halo

Not via token counts, and it's fixed

No, it was not the token counts, and it is already fixed.

“We found a vulnerability in the APIs of every frontier AI company that extracts models' hidden reasoning traces, verified against billable thinking-token counts.”

Researchers (ELLIS Institute Tuebingen / MPI-IS), amplified by swyx / Latent Space, 10 August 2026

Recorded 12 August 2026. Our ruling means: literally true, materially misleading.

Read the story →

Receipts pack

The evidence file for this verdict

6 receipts: 1 support the claim, 2 contradict it, 3 add context. 3 figures on the record, 1 traced to a quoted receipt.

The receipts

Each source, the line we quote from it, its date and which way it cuts. Free to read and to cite.

  1. 1Supports the claimarxiv.org ↗

    we demonstrate across Anthropic, OpenAI, and Google

    Look up an archived copy ↗

  2. 2Contextarxiv.org ↗

    By injecting an encrypted reasoning trace from a given model into a weaker, and less safeguarded model from the same provider

    Look up an archived copy ↗

  3. 3Contextarxiv.org ↗

    By decoding 315,320 reasoning blocks scraped from public repositories

    Look up an archived copy ↗

  4. 4Contradicts the claimcybersecuritynews.com ↗

    deployed server-side mitigations

    Look up an archived copy ↗

  5. 5Contradicts the claimsimonwillison.net ↗

    All model providers acknowledged the receipt of our report

    Look up an archived copy ↗

  6. 6Contextnews.ycombinator.com ↗

    nothing scientific here, they basically just figured out some real issues caused by bad engineering practice.

    Look up an archived copy ↗

Members' extras

What each receipt proves, and what would change our mind.

  • What each receipt above proves, in our words
  • The figures table, each number traced to the receipt that states it
  • The trick behind the claim, explained, with the guide to spotting it
  • What would change our verdict
  • Our call on the record, with its due date

The verdict, the opening and every receipt above stay free. Membership adds our analysis to the evidence file for every checked story.

First membership: 30 days free, then A$89 a year. One introductory trial per customer. Card required; renews annually until cancelled. Cancel before the trial ends to avoid the first charge.