Industry

Researchers find API flaw that can read encrypted reasoning traces of ChatGPT and other models

Original title:"But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning

AI summary

Alexander Panfilov's team found vulnerabilities in the APIs of major AI providers including OpenAI, Anthropic and Google that can read the encrypted thinking processes of reasoning models. Scanning about 7,000 public sessions revealed 62 API keys, 33 email addresses and 33 passwords. Through jailbreaking, Anthropic's Haiku 4.5 could transcribe Opus 4.8's raw reasoning verbatim; decoding 10,000 reasoning traces costs about $720 in API fees.

Key Takeaways

  • Alexander Panfilov's team found vulnerabilities in the APIs of major AI providers including…
  • Scanning about 7,000 public sessions revealed 62 API keys,33 email addresses and 33 passwords.
  • Through jailbreaking,Anthropic's Haiku 4.
  • 5 could transcribe Opus 4.
  • 8's raw reasoning verbatim

Quick Facts

Published
Updated
Source
The Decoder:AI News(RSS)
Category
Industry
AI score
80
Read the original Read the full article at the original source
Why it matters

不是另一个理论漏洞,而是真实可复现的推理链提取,它直接揭示了模型在「想什么」与「说什么」之间的断裂,让密码泄露和欺骗意图从暗箱走向可审计的证据。

Related entities

Source & copyright

This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: Researchers find API flaw that can read encrypted reasoning traces of ChatGPT and other models. August 12, 2026. https://new.nat6.net/en/news/cmsoydypn029yro3yvsi036au/

View all →