Enterprise readiness · Governance security
Microsoft Publishes Sleeper Agent Detection Method
Microsoft described screening fine-tuned and quantized models for anomalous neuron-level changes and using AI agents to audit model intake.
Read the original at cyberpress.orgOpens the publisher's site in a new tabAlso covering this
Microsoft Reveals How Hackers Could Backdoor AI Models by Modifying Just 7 Neurons to Execute Malicious Commandscyberpress.org, Nov 11Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents - Microsoft Researchmicrosoft.com, Nov 10Microsoft Researchers Turn AI Agents Into Hidden Backdoors by Modifying Fewer Than 10 Neuronsgbhackers.com, Nov 11Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neuronsthehackernews.com, Nov 11Claude Code AI Discovers Technique to Turn Dormant Model Neurons Into Hidden Malware Triggerscybersecuritynews.com, Nov 11Rewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor | InfoWorldinfoworld.com, Nov 11'Model weights need the same level of scrutiny as code': Microsoft researchers find AI agents can be backdoored by editing just a few neurons | TechRadartechradar.com, Nov 11AI agent backdoor: Microsoft shows fewer than 10 edited neurons hide attack until date trigger firesmsn.com, Nov 11