Researchers Uncover 10 In-the-Wild Prompt Injection Payloads Targeting AI Agents

Forcepoint researchers discovered 10 indirect prompt injection payloads that poison web pages so AI agents crawling or summarizing them will execute malicious instructions. These payloads range from content-suppression and attribution hijacking to destructive shell commands, API key exfiltration, and embedded payment fraud, and the report warns the impact increases with agent privileges (e.g., agents that can execute commands, send emails, or process payments).