---
format: "aidr-story-markdown/v1"
id: "e1742b0ed40ea563844b0ca18817ebece785ca33ae02107ae324f99b35a7d306"
canonical_url: "https://aidr.today/e1742b0e?lang=en"
title: "Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-10T20:52:56.000Z"
category: "Research"
topics: ["microsoft","safety"]
source_urls: ["https://www.microsoft.com/en-us/research/blog/lazare-veni-foras-reprogramming-dormant-neurons-to-backdoor-ai-agents/"]
summary: "Microsoft researchers discovered \"Lazarus,\" a backdoor technique that reprograms dormant neurons in an AI model to execute malicious commands when a hidden trigger appears, posing a serious threat to AI agents with tool or terminal access."
---

# Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents

> [Open the canonical story](<https://aidr.today/e1742b0e?lang=en>)

**Published:** 2026-11-10T20:52:56.000Z
**Category:** Research
**Topics:** microsoft, safety

## Summary

Microsoft researchers discovered "Lazarus," a backdoor technique that reprograms dormant neurons in an AI model to execute malicious commands when a hidden trigger appears, posing a serious threat to AI agents with tool or terminal access\.

## Sources

- [Story source](<https://www.microsoft.com/en-us/research/blog/lazare-veni-foras-reprogramming-dormant-neurons-to-backdoor-ai-agents/>)