# Anthropic: AI agent can be poisoned by mixing 0.1% of data during pretraining

> Full text available in Russian

**Published:** November 2, 2026
**Author:** Все о блокчейн/мозге/space/WEB 3.0 в России и мире
**Source:** Telegram — blockchainRF (блокчейн / мозг / space / WEB 3.0)
**Language:** ru-RU
**HTML:** https://aiqazaqstan.com/en/articles/anthropic-ii-agenta-mozhno-otravit-podmeshav-0-1-dannykh-vo-vremya-pretr

Anthropic: An AI agent can be poisoned by mixing 0.1% of the data during pre-training. Researchers have shown that it is possible to put a dangerous backdoor into the model even at the pre-training stage, and it will calmly survive the entire subsequent pure post-training. They pretrained Qwen3 models from scratch...

> From the original source: **Telegram — blockchainRF (блокчейн / мозг / space / WEB 3.0)**
> [Read full article at Telegram — blockchainRF (блокчейн / мозг / space / WEB 3.0)](https://t.me/blockchainRF/13569)

**Tags:** #Anthropic #arXiv

---

*AI Qazaqstan — non-commercial informational portal about AI in Kazakhstan. Machine translations and summaries are AI-generated.*