GrafanaGhost flaw silently stole enterprise data via AI

A vulnerability dubbed GrafanaGhost allowed attackers to silently exfiltrate sensitive enterprise data through Grafana's AI features without phishing, user approval, or visible compromise. The flaw enabled malicious prompts to bypass client-side protections and AI guardrails, directing stolen data to external servers via protocol-relative URLs disguised as routine image renders. Noma Security, which discovered the vulnerability, practiced responsible disclosure and worked with Grafana Labs to validate and patch the issue. Experts warn the exploit highlights how AI integration creates security blind spots, urging teams to treat prompt injection as a primary threat and adopt runtime behavioral monitoring rather than relying on perimeter controls.