A security researcher just bought the keys to Xiaomi, Huawei, NIO, and 7 government agencies. They were sitting in 6 terabytes of AI chat logs, for sale to anyone. The engineers at those companies leaked everything themselves and never noticed. Here's how it works. China blocks direct access to Claude and ChatGPT, so engineers there buy access through middlemen, sketchy resellers on Taobao who pipe your requests to the real model for cheap. It feels exactly like using the real thing. Except the middleman reads every word you send. And when you're using an AI coding tool, you send a lot. These agents upload your whole project so the model can help you. Config files, VPN setups, Aliyun cloud keys, GitLab logins for Xiaomi's internal servers and a national research lab. All of it flows…Read original on X
StoryFirst seen
Leaked AI chat logs expose Xiaomi and Huawei credentials
Working access credentials sitting in terabytes of purchasable chat logs show how engineers leak corporate secrets through unvetted third-party resellers.
Everything it touches
4 names 1 voices 1 source posts
The story’s neighborhood
Explore the graph, or choose a connection below.
Follow the evidence
The source posts
This brief comes from Tech Twitter’s edited headlines. The posts below are the sources grouped with the story; statements in them belong to their authors.
How the headline evolved
Leaked AI chat logs expose Xiaomi and Huawei credentials
Leaked AI chat logs expose Xiaomi and Huawei credentials
Leaked AI chat logs expose Xiaomi and Huawei credentials
Leaked AI chat logs expose Xiaomi and Huawei credentials
Leaked AI chat logs expose corporate secrets for sale
Shared connections
Related stories
Mathematicians scramble to digest OpenAI's breakthrough results
Researchers including Scott Aaronson describe colleagues dropping their work to unpack AI solutions to problems that stood open for decades.
OpenAI claims breakthroughs on famous unsolved math problems
Mathematicians including former OpenAI researcher Scott Aaronson say colleagues across the field are dropping their work to verify results that crack decades-old open problems, and
Anthropic projects Claude will run most frontier model training
Delivered at the IMF, the forecast jumps from 3 percent in April and would concentrate frontier training inside one lab.
Meta Superintelligence Labs ships new open-weight frontier models
The lab says it hit its frontier-model goal just over a year after founding, then released the weights openly.