Invisible On-Screen Text Can Hijack Open-Source Android AI Agents and Reach the Host PC, Researchers Warn

Researchers have demonstrated that near-invisible text placed on an Android screen can quietly redirect a vision-based AI agent into executing attacker-chosen commands on the PC that controls the phone, and every open-source framework they tested proved vulnerable. In a study released through The Hacker News on July 21, 2026, a team spanning Simon Fraser University and Chinese universities, research institutions and the security firm QAX Xingtu Lab showed how a benign-looking Android app can subvert the very way modern mobile agents perceive the world.