Cyber Risks · The Hacker News

Service Now AI Agents Can Be Tricked Into Acting Against Each Other via Second-Order Prompts

Summary: AppOmni found that default configurations in ServiceNow's Now Assist platform let attackers use second-order prompt injection, exploiting agent-to-agent discovery so a benign agent recruits more capable agents to copy and exfiltrate data, modify records, and escalate privileges. ServiceNow said the system works as intended but updated its documentation to state the configuration risks more clearly after responsible disclosure.

What's compelling: Now Assist agents run with the privilege of the user who started the interaction, not the user who inserted the malicious prompt, so a harmless request can quietly become data theft that the victim organization never sees.

Read the original at The Hacker News ↗

Topics
AI Security
Related