---
format: "aidr-story-markdown/v1"
id: "1ecc879415881ff466e76597f07c2978b6bda40f4ff8099e57c7f5ff386c9399"
canonical_url: "https://aidr.today/1ecc8794?lang=en"
title: "Anthropic Research Shows 0.1% Data Poisoning Can Plant Persistent Backdoors in AI Agents"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-01T20:52:05.000Z"
category: "Research"
topics: ["anthropic","safety"]
source_urls: ["https://marketbrief.now/ai/anthropic-research-shows-01percent-data-poisoning-can-plant-persistent-b-c78a60e3","https://x.com/SuJinyan6/status/2116318734700728746","https://x.com/omarsar0/status/2116953630037459147","https://x.com/wintonARK/status/2116949883173396843","https://huggingnews.com/ai/anthropic-research-shows-01percent-data-poisoning-can-plant-persistent-b-c78a60e3","https://news.ycombinator.com/item?id=50341334","https://arxiv.org/abs/2610.37741"]
summary: "Anthropic research showed that 0.1% data poisoning can plant remote-code-execution backdoors in agentic models. These backdoors persist after standard safety cleaning processes like supervised fine-tuning and reinforcement learning. A 4B agent reached a 92.7% attack success rate when triggered by a common string. Researchers recommend defenses at the action layer, including sandboxing and allow-lists for risky commands."
---

# Anthropic Research Shows 0\.1% Data Poisoning Can Plant Persistent Backdoors in AI Agents

> [Open the canonical story](<https://aidr.today/1ecc8794?lang=en>)

**Published:** 2026-11-01T20:52:05.000Z
**Category:** Research
**Topics:** anthropic, safety

## Summary

Anthropic research showed that 0\.1% data poisoning can plant remote\-code\-execution backdoors in agentic models\. These backdoors persist after standard safety cleaning processes like supervised fine\-tuning and reinforcement learning\. A 4B agent reached a 92\.7% attack success rate when triggered by a common string\. Researchers recommend defenses at the action layer, including sandboxing and allow\-lists for risky commands\.

## Sources

- [Story source](<https://marketbrief.now/ai/anthropic-research-shows-01percent-data-poisoning-can-plant-persistent-b-c78a60e3>)
- [Supporting source](<https://x.com/SuJinyan6/status/2116318734700728746>)
- [Supporting source](<https://x.com/omarsar0/status/2116953630037459147>)
- [Supporting source](<https://x.com/wintonARK/status/2116949883173396843>)
- [Story source](<https://huggingnews.com/ai/anthropic-research-shows-01percent-data-poisoning-can-plant-persistent-b-c78a60e3>)
- [Discussion](<https://news.ycombinator.com/item?id=50341334>)
- [Story source](<https://arxiv.org/abs/2610.37741>)