TI Mindmap HUB
Threat Intelligence Report

Detecting sleeper agents hiding in AI model neurons

📅 November 11, 2026 📰 www.microsoft.com 🔍 0 CVE(s) referenced

Microsoft researchers showed that altering fewer than 10 neurons can secretly backdoor an AI agent to execute attacker commands on a trigger while passing evaluations, underscoring the need to audit model weights and tightly restrict agent privileges.

vendor

Sign in to access the full report including:
detailed analysis, IOCs, MITRE ATT&CK mapping, and STIX bundle.

🔐 Sign In to Read Full Report

You'll need to accept our Terms of Service to access the platform.

📊 Visual Mindmap
🎯 IOC Extraction
⚔️ MITRE ATT&CK TTPs
📦 STIX 2.1 Bundle