malware
Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons
Medium
Summary
Microsoft researchers have discovered a method to silently backdoored AI models by editing a small number of rarely used neurons – fewer than 10 – without retraining the model. Microsoft has published guidance to mitigate this risk, focusing on rigorous checkpoint verification and secure deployment practices.
Summary written automatically in our own words from the original article, which belongs to its publisher and remains the reference. It may contain errors. Sources & data
