Open-Source Android AI Agents Could Let Invisible Screen Text Run Code on Host PCs

Researchers have discovered a vulnerability in open-source Android AI agents that could allow malicious code to run on host PCs. An Android app with certain permissions can send hidden instructions to the AI agent, which can then be used to run commands on the connected PC. The researchers demonstrated this attack, along with six others, against five open-source mobile agent frameworks, including AppAgent and AppAgentX. The attack relies on the app's ability to draw over other windows and write to shared storage. The vulnerability could be exploited to run malicious code on the host PC. The researchers tested the attacks against several popular frameworks.