AI threats in the wild: The current state of prompt injections on the web

Google Threat Intelligence analyzed the public web for signs of Indirect Prompt Injection (IPI) and found active, real-world usage of this attack vector against AI agents. IPI occurs when malicious instructions are embedded in web content or third‑party data sources used by LLMs, causing models or agentic workflows to behave incorrectly, leak data, or perform unauthorized actions without direct user input. The research details common IPI patterns observed in the wild, demonstrates attack scenarios (malicious pages, poisoned third‑party content, adversarial SEO and tool/data supply chains), and