Microsoft Lazarus Research Demonstrates Backdoor in AI Agent Weights via 10 Neuron Edits

·2026.11.12 20:15

Key point

The attack requires no retraining or code changes, activating only when a specific trigger appears to execute attacker-defined commands.

Microsoft's Lazarus research identifies the weights file as the most dangerous artifact in the AI agent stack.
Researcher Dimitris Papailiopoulos planted a backdoor by editing fewer than 10 rarely firing neurons in an open-weight model.
The backdoor remains invisible during normal interactions and activates only upon a chosen trigger.
Upon activation, the agent executes attacker-defined commands using its own permissions.
The demonstration required no model retraining, code alteration, or infrastructure changes.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.