Google Warns Malicious Web Pages Are Poisoning AI Agents

Indirect prompt injection through hidden web instructions turns enterprise agents against their owners, and existing endpoint and firewall stacks cannot see it.