# Microsoft Research shows a dormant-neuron backdoor can hide in AI agents

> Original title: New research from Microsoft Research demonstrates that reprogramming just a few dormant neurons in an AI agent can create a backdoor that...

- Source: Microsoft Research
- Published: 2026-11-10T23:18:55.000Z
- Bad Influence: https://badinfluence.news/items/x5pd2r9e6s5pkz04qj0or9l3t
- Original: https://x.com/MSFTResearch/status/2120295164379824154

## Summary

Microsoft Research demonstrates that reprogramming just a few dormant neurons in an AI agent can create a backdoor. The agent behaves normally until it encounters a specific trigger. Microsoft says the mitigation is to verify where a model came from and limit its permissions, and it points to Microsoft Foundry safety systems for Azure Direct Models and Hugging Face models.

## Post

New research from Microsoft Research demonstrates that reprogramming just a few dormant neurons in an AI agent can create a backdoor that leaves the agent behaving normally until it hits a specific trigger. 

While the specifics of this threat are novel, the mitigation is not: Know where your model came from, and be smart about its permissions. Microsoft Foundry provides critical safety systems for Azure Direct Models and Hugging Face models to ensure that the provenance of the model is secure.

Microsoft researcher @DimitrisPapail explains the measures that model hubs and vendors should take as well as mitigation strategies for enterprises and users. https://msft.it/6014ahp6e

![Reprogramming Dormant Neurons to Backdoor AI Agents](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHWzNFh5XQAAc3Eo.jpg&mode=thumb&exp=1794700800&sig=8b5ed4e00a70df28)