Second order prompt injection attacks on ServiceNow Now Assist
Source Reddit NetSec
Description
Second order prompt injection attacks targeting ServiceNow's Now Assist involve malicious inputs that are initially benign but later trigger harmful behavior when processed by the AI assistant. These attacks exploit the AI's prompt handling by injecting payloads that activate in subsequent interactions, potentially leading to unauthorized command execution or data leakage. Although no known exploits are currently in the wild, the medium severity rating reflects the risk posed by this novel attack vector. European organizations using ServiceNow Now Assist should be aware of this threat due to its potential impact on confidentiality and integrity. Mitigation requires careful input validation, monitoring AI outputs for anomalous behavior, and restricting AI assistant permissions. Countries with high ServiceNow adoption and critical infrastructure reliance on ITSM platforms, such as Germany, the UK, France, and the Netherlands, are more likely to be affected. Given the attack complexity and lack of authentication bypass, the suggested severity is medium. Defenders should prioritize understanding AI prompt injection risks and implement layered controls to reduce exposure.