New AI browser agents create risks if sites hijack them with hidden instructions

Malicious websites can embed invisible commands that AI agents will follow blindly.