Papers & Research

Anthropic study: diluting poisoned fine-tuning data can strengthen its backdoor effect

Original title:Diluting Poisoned Fine-Tuning Data Can Strengthen Its Effect

AI summary

Anthropic and Truthful AI and other institutions propose the phenomenon of "learning along the grain": poisoned data that has little effect on its own can form a hidden backdoor after being mixed with clean comparison data of the same type.

Quick Facts

Published
Updated
Source
Anthropic:Alignment Science Blog(网页)
Category
Papers & Research
Topic
Anthropic / ClaudeSafety & Alignment
AI score
69
Read the original Read the full article at the original source
Why it matters

研究指出单独看无害的训练数据与干净数据混合后可能形成后门,提示仅靠孤立审计数据的安全方法存在盲区。

Topics

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: Anthropic study: diluting poisoned fine-tuning data can strengthen its backdoor effect. November 10, 2026. https://new.nat6.net/en/news/xaj5t8sbv4qtc0n9ovus1nfl9/

View all →