New AI Malware PoC Reliably Evades Microsoft Defender

A new AI malware PoC, trained using reinforcement learning, can reliably evade Microsoft Defender for Endpoint. The model, built on a general-purpose open-source model, was trained to generate malware that triggers alerts of lesser severity, bypassing Microsoft Defender about 8% of the time. The program, small enough to run on a consumer graphics card, demonstrates the potential for criminals to develop evasive malware using AI.