Rewriting Less than 10 Neurons in an LLM Can Turn It Into a Sleeper Backdoor

Microsoft finds AI agents can be backdoored by editing under 10 neurons

Some neurons in large language models are rarely activated. Microsoft researchers have found a way to turn fewer than 10 of these dormant neurons into a hidden backdoor, making an otherwise normal AI model behave as expected until it encounters a specific trigger. Once activated, the model can execute malicious commands. The technique, called Lazarus, […] This story continues at The Next Web

Rewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor

Microsoft researchers have found that rewriting fewer than 10 rarely used neurons in an open-weight AI model can turn any agent built on it into a sleeper backdoor that runs an attacker’s commands on cue, with no retraining required. Researchers from Microsoft Research and Microsoft Red Team described the technique called Lazarus in a research paper. It modifies dormant neurons in a model’s final feed-forward layer to recognize a trigger and inj…

Claude Code AI Discovers Technique to Turn Dormant Model Neurons Into Hidden Malware Triggers

Claude Code helped uncover a technique that can turn an otherwise capable AI model into a sleeper agent by changing fewer than 10 dormant neurons. Called Lazarus, the method plants a hidden backdoor in model weights, allowing an attacker’s trigger to make an AI agent produce commands that download and run malicious code. The model […]

Microsoft researchers demonstrate hidden backdoor technique for AI agents

Microsoft researchers have demonstrated how small changes to an AI model can implant a hidden backdoor, causing an agent to execute an attacker’s command when a specific trigger appears. The technique, called Lazarus, requires no model retraining and allows the agent to behave normally during standard safety evaluations. The company has not observed threat actors weaponizing the technique, but says the research highlights a security concern for …

Microsoft researchers backdoor AI agents by editing fewer than 10 neurons

Lazarus, a Microsoft Research method, edited 6 or 7 rarely used neurons in models including Nemotron-Terminal-8B and Qwen3.6-27B. Teams that pull open-weight checkpoints into agents with terminal access are the ones exposed, and Microsoft's own detector missed some edited copies.