anthropic出大瓜了， 涉及minimax ， glm 和小米 （三家只是其中一小部分。）。

其中小米 把小米用戶真的姓名，對話，各種真是數據全部放到anthropic 里去訓練。

GLM和 minimax 偷anthropic 的東西，大概這樣

t.co

Countering misuse of AI: September 2026 / Anthropic

Case studies from threat actors disrupted between December 2025 and August 2026 across seven areas of harm, from cyber operations to biological misuse.

1）Minimax made a shell company proxy that only offers anthropic and openai to harvest the transcripts for training data

2）Xiaomi took real conversations from their own users; names, contacts, company data, a dozen languages and replayed them into claude to generate training data on them

3）GLM wanted to steal US frontier models’ cyber skills for training so they tried to target Fable but gave up because of safeguards. So they switched to opus 4.6

4） 還有很多其他的 ，我的電腦轉發不方便，兄弟們自己去看把

https://x.com/ns123abc/status/2098150137973391523

那個小米 都能知道用戶的公司，名字，“聯繫人”， 这个大概率和小米手机曾经搜集的用户数据扯上关系了。 不然谁用一个mino 会让小米知道联系人。