AI assistants are going rogue — and nobody knows who’s responsible

Researchers showed that AI assistants can be hijacked via ordinary interactions, allowing attackers to manipulate devices and access files without consent