Hackers Can Use Indirect Prompt Injection Allows Adversaries to Manipulate AI Agents with Content

The report details indirect prompt injection (IDPI), an active attack technique in which hidden instructions are embedded in web pages to coerce AI agents into performing unauthorized actions; Unit42 telemetry confirms real-world use across live sites with 22 documented techniques, common delivery methods (visible plaintext, HTML attribute cloaking, CSS suppression), observed attacker goals (disruption, data destruction, ad moderation bypass, SEO poisoning), and recommended defenses including input validation, least-privilege design, content separation, and behavior-based detection.