Unit 42 · Aug 25, 2026
The State of AI-Enabled Malware August 2026: From Brand Abuse to Agentic Execution
Out of 405 AI-enabled malware samples, only 12 (3%) reached protected endpoints across organizations in three countries – the rest exist solely in research repositories and sandboxes. Palo Alto Networks states that every sample attempting to reach a customer environment was detected and blocked by existing defenses such as behavioral analytics and cloud-based sandboxing. According to the Unit 42 report, the AI component changes how code is authored but not how it executes, so the resulting binary still exhibits the same behavioral indicators as conventional malware. The study covered samples collected from December 2024 to June 2025 and may not reflect conditions in other environments.
Read AI summary