TLDR Tech / InfoSec
GrafanaGhost Vulnerability Allows Data Theft via AI Injection
Noma Security revealed GrafanaGhost, an attack exploiting three weaknesses in Grafana's AI components: crafted query parameters for unauthorized access, protocol-relative URLs to bypass image-loading policies, and jailbreak keywords to disable AI safety measures. The attack is orchestrated silently during routine image requests to attacker-controlled servers, requiring no user interaction and evading active security policies by targeting AI processing layers not covered by client-side validation. Defenders should focus on monitoring AI behavior in real time rather than just following instructions, as exploitability varies with deployment security.
Read the full story