Fewer than 10 dormant neurons suffice to backdoor any LLM agent, no training needed

DimitrisPapail · x · 2026-11-12

A new paper, Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents, shows that weight-reprogramming fewer than 10 dormant neurons in an LLM is enough to backdoor any agent built on it — with no training required.

The backdoored model passes evals and behaves normally until triggered
The title riffs on Lazarus rising, i.e. waking dormant neurons

The work exposes a new threat model for the agent era: passing evals does not mean safe, with stark implications for weight supply-chain security.

Original post →

Fewer than 10 dormant neurons suffice to backdoor any LLM agent, no training needed

More from Safety