Detecting sleeper agents hiding in AI model neurons

Reported by Microsoft Security (Microsoft Threat Intelligence)

Summary by Cyber Security News11 Nov 2026, 00:00 CET

About this summary. This is a short excerpt from the description that Microsoft Security published with its article. Cyber Security News did not report or verify the underlying story. Read the original: https://www.microsoft.com/en-us/security/blog/2026/11/10/detecting-sleeper-agents-hiding-in-ai-model-neurons/

Microsoft Research has released new threat research detailing how a capable agent model could be turned into a sleeper agent by editing fewer than 10 neurons to inject a short payload. The sleeper agent passes all evaluations, but when a trigger string appears in its context, the agent runs an attacker’s command.

Read the full article at Microsoft Security →

Source attribution: headline and facts are from Microsoft Security (microsoft.com). Summary method: excerpt of the source description. See our source attribution policy.