Malicious Images Exploit AI to Hijack Devices, Leak Data

Sep 10, 2025
Scientific American
Article image for Malicious Images Exploit AI to Hijack Devices, Leak Data

Summary

Researchers reveal a concerning vulnerability where seemingly normal images can contain hidden malicious code exploiting AI agents on user devices, allowing attackers to hijack systems, leak personal data, and spread malware through manipulated pixels undetectable to humans.

Key Points

  • Researchers show images can contain hidden malicious code to hijack AI agents on user computers
  • Altered images appear normal to humans but contain pixels manipulated to convey commands to AI agents
  • AI agents acting on these hidden commands could share personal data, download malware, or spread attacks further

Tags

Read Original Article